As AI adoption grows, so does organizations’ appetite for the vast data from disparate sources needed to train AI models. Because of this, companies are grappling with how to safeguard a surging amount of fragmented data wherever it lives.
The post DSPM: A Cybersecurity Approach Tailor-Made for This AI Era appeared first on Security Boulevard.
A critical security vulnerability in Rockwell Automation’s ControlLogix and GuardLogix controllers has been discovered. This vulnerability could potentially allow attackers to bypass security measures and gain unauthorized access to industrial control systems. Researchers at Claroty’s Team82 uncovered the flaw, which is identified as CVE-2024-6242. It affected various models of Rockwell’s 1756 ControlLogix devices. The vulnerability, […]
The post Rockwell Automation Devices Flaw Let Hackers Gain Unauthorized Access appeared first on Cyber Security News.
LLMs are different from other tools and different approaches are required to mitigate their risks involving new security technologies.
The post Strategies for Mitigating LLM Risks in Cybersecurity appeared first on Security Boulevard.
A newly discovered vulnerability in Windows File Explorer has raised alarms within the cybersecurity community. Identified as CVE-2024-38100, this security flaw allows attackers to escalate privileges by exploiting a seemingly innocuous wallpaper feature. CVE-2024-38100 – Windows File Explorer Elevation of Privilege Vulnerability Released on July 9, 2024, CVE-2024-38100 is a critical vulnerability that Microsoft has […]
The post Leaked Wallpaper Vulnerability Exposes Windows Users to Privilege Escalation Attacks appeared first on GBHackers on Security | #1 Globally Trusted Cyber Security News Platform.
The communication between DNS recursive resolvers and authoritative nameservers is largely unsecured, making it susceptible to on-path and off-path attacks. Though many security proposals have been put forward, they often face implementation challenges or lack adequate security features. This persistent vulnerability reveals the need for a new, widely deployable secure scheme that will overcome the […]
The post DNSSEC+ – Secure Model That Addresses Security And Downsides Of DNSSEC appeared first on Cyber Security News.
Cybersecurity researcher Jeremiah Fowler discovered and reported to VpnMentor about 13 non-password-protected databases containing 4.6 million documents, including sensitive voter records and election-related documents. This breach raises significant concerns about data protection and the security of election systems in the United States. Discovery of the Breach Jeremiah Fowler’s investigation began when he stumbled upon a […]
The post 4.6 Million Voter Database & Election Documents Exposed Online appeared first on Cyber Security News.
Ransomware operators often acquire malware through purchases on the dark web, group affiliations, and leaked source codes rather than developing themselves. They target victims by using common tools and modified samples to propagate attacks. Recent reports by the security analysts at Kaspersky Lab suggest that new emerging groups like SEXi utilize different leaked ransomware variants, […]
The post Threat Actor Groups Using Leaked Ransomware Variants To Launch Attacks appeared first on Cyber Security News.
The Payment Card Industry Data Security Standard (PCI DSS) aims to improve credit, debit and cash card transaction security and protect cardholders from breaches of their personal information.
The post Effective Third-Party Risk Management Under PCI DSS 4.0 appeared first on Security Boulevard.
8月1日，由360集团AI安全实验室、中国科大网络空间安全学院主办的人工智能安全趋势发展论坛在京圆满落幕。本次论坛汇聚了人工智能领域政府、学术界、产业界的代表和优秀专家学者，就人工智能安全挑战、最新研究成果及最佳实践进行深入交流，共同推动形成行业共识与标准，推动人工智能安全技术的创新与发展。
图 ISC人工智能安全趋势发展论坛现场
大模型已成为数智化系统的关键基础设施，其潜在的主体地位和应用广度，也带来了新的安全风险和安全挑战。360集团智脑总裁张向征在致辞指出，人工智能安全与发展的平衡，不仅需要法律法规的约束，更需要行业内各方力量共同推动和构建人工智能安全生态。会上，360作为数字安全领军企业，基于“安全、向善、可信、可控”安全四原则构建了系统化的大模型安全解决方案，通过技术革新，实现大模型风控效率提升100倍，幻觉比例下降50%，为全球大模型安全赋能；而运用全新技术路线构筑“看见—处置”安全大模型，创新性打造类脑框架，五大中枢可以类比人类大脑实现专家级安全能力，为企业打造7*24小时高级安全专家团队。
图 中国科学技术大学网络空间安全学院执行院长俞能海
中国科学技术大学网络空间安全学院执行院长俞能海教授在报告中指出，人工智能大模型安全问题的研究方兴未艾，有待于广大的学者和行业人士共同努力，为国家提供更多成熟可用的产品，能够真正地服务千行百业。此外，他还在报告中概述了生成式人工智能的发展脉络和安全风险，重点分享了人工智能大模型安全在数据采集与处理、模型训练与测试、内容生成与服务几个关键阶段的最新研究进展。
迈入大模型时代，更大规模的数据以及更强的任务处理能力，也带来了更大的安全与隐私风险。西安交通大学二级教授、副处长沈超认为，大模型能够更安全、可持续的发展，需要去关注大模型在完整性、可靠性方面存在的问题，以及在数据模型应用方面存在的安全隐私挑战。
图 中国信通院人工智能研究所安全部主任石霖
大模型带来生成方式的转变，推动生产领域交互体验革新和效率提升。中国信通院人工智能研究所安全部主任石霖表示，人工智能安全问题涌现，持续放大数字社会风险与挑战，信通院基于已有的安全评估测试等工作基础，设计了大模型价值对齐方法，对开源大模型自身进行安全防护加固，获得了明显的安全性能提升。
图 北京阿莱门科技有限公司CEO徐骅
北京阿莱门科技有限公司CEO、北京大学人工智能安全与治理中心大模型研究员、北大(临港)大模型对齐中心执行主任徐骅指出，AI安全成为大模型广泛应用的关键掣肘因素，价值对齐是AI安全的核心议题，也是大模型应用的最后一公里。为此，阿莱门推出覆盖大模型全生命周期的安全对齐方案，提供全方位的大模型安全对齐平台，提升大模型的安全性和能力。
图 华为可信AI安全解决方案专家唐文
AI创新发展与监管并重，华为可信AI安全解决方案专家唐文认为，安全可信的AI计算平台才能满足市场准入与监管合规的要求。华为可信业界首创“用AI保护AI”，依托昇腾AI算力平台，实现了物理对抗攻击检测和大语言模型（代码）提示注入攻击检测。
论坛最后，各位专家学者围绕政企研的角色定位、数据合规、幻觉现象、安全评测、安全和性能平衡、大模型安全领域的发展趋势等内容进行了圆桌研讨和交流。此次人工智能安全趋势发展论坛的举办，政府、学术界、产业界将共同努力，为构建一个更加稳健、更加透明、更加负责任的人工智能生态系统贡献力量。