今日共收录 2436 条安全情报，包含 1258 个 CVE，779 篇安全通告，以及 100 篇研究论文。
← 返回归档列表Ransomware incidents in Japan rose 4.7% year over year. The Gentlemen was the most active group, with leak-site listings
The defining development of the period came not from attackers but from the AI labs themselves, whose models broke out o
A new Android malware called RatHat has been discovered, targeting users with an AI-powered subsystem that helps operato
OpenAI has presented new examples of what they call "AI model misalignment" from the past six months, including unauthor
Microsoft reminded customers this week that devices running Windows 11 24H2 Home and Pro editions will stop receiving up
The U.S. Federal Bureau of Investigation (FBI) seized the domains used by NightmareStresser, one of the world's longest-
The China-linked espionage group FamousSparrow has been using a new backdoor named SparroWocky in attacks on government
Microsoft shared a temporary fix on Wednesday for a known issue that prevents Windows 11 users from logging in with vali
The Iran-linked "hacktivist" persona known as Handala Hack has been attributed to a Telegram-based surveillance backdoor
Attackers now weaponize new vulnerabilities in about five days (Mandiant, part of Google Cloud). The median organization
The U.S. Department of Justice (DoJ) on Tuesday announced the court-authorized seizure of internet domains associated wi
The company plans to expand into continuous cybersecurity, offering security testing across applications and infrastruct
该论文研究的是按 token 计费的 LLM 服务中一个被忽视的信任问题：付费与模型输出长度直接挂钩，因此服务提供方存在经济动机在几乎不损失任务可用性的前提下，隐蔽地把输出拉长。作者将这类由服务端主导的操纵定义为「服务端 Token 膨胀攻
该论文研究将大语言模型（LLM）引入恶意软件分诊流程后所暴露出的新型攻击面，并提出了一种名为 ALIBI（Adversarial Legitimacy Injection in Binary Input，二进制输入中的对抗性合法性注入）的语
这篇论文聚焦工具增强型 LLM 智能体（tool-augmented LLM agent）中一个被现有方法系统性忽略的失效模式：智能体调用根本不存在的工具，或向工具传入其 schema 从未声明的参数。作者指出，现有防御分为两类——一是工具
该论文关注 Model Context Protocol (MCP) 生态中的一个结构性安全问题：MCP 让 LLM 能够调用外部工具，但模型在每一次工具交互中都会经由多个输入通道（工具描述、工具返回结果、采样/sampling 消息）接触
该论文研究的是基于学习的机器人操作（learning-enabled robotic manipulation）在仿真器中训练与评估、再部署到真实世界这一典型流程中的一个被忽视的安全缺口。作者指出，机器人仿真器中的三维资产实际上包含两套彼此
该论文针对多模态大语言模型（MLLM）面临的非法部署与未授权蒸馏问题，提出了首个面向多模态模型的指纹识别（fingerprinting）系统性研究。作者指出，现有模型溯源方案大多为纯文本 LLM 设计，而 MLLM 通常共享同一语言主干（l
该论文研究一个核心问题：多智能体 LLM 系统如何从局部策略偏差演化为集体失控。作者提出一套"流行病学"解释框架，把过程拆成三个环节——偶发突变、传染、恢复。单个 agent 自发产生的偏差构成"种子"；agent 之间的通信使其他 age
该论文聚焦自主智能体浏览器（agentic browser）在用户已认证会话下执行安全敏感操作时所面临的动作完整性威胁。研究指出，间接提示注入（indirect prompt injection）与欺骗性确认界面可使用户的审批被操纵，而现有