Security News
全球顶尖安全社区实时聚合，追踪最新漏洞、攻防技术与行业动态。
来源：先知社区 · Seebug · The Hacker News 等权威平台 | 每日自动更新
RAG投毒攻防--从检索污染到激活检测
复现了PoisonedRAG，顺便把整个攻击链和检测机制拆了一遍，记录一下
当 AI 开始“代你执行”：企业级 AI Agent 的真实攻击面与安全边界重构
随着 AI 从“信息生成工具”走向“可执行系统”，其在企业环境中的角色正在发生本质变化。越来越多的 AI Agent 被接入工单系统、自动化运维、云资源管理与内部业务编排流程，开始直接参与决策与操作执行。 在这一背景下，传统基于身份认证与权限校验的安全模型，正在被自然语言上下文、状态记忆与推理结果所 ...
ObliInjection：针对多源数据 LLM 智能体的顺序无关提示注入攻击
前言在现实应用中，许多 LLM 智能体（如智能客服、AI 助手、自动化代理）会从多个独立数据源（例如用户输入、数据库查询结果、网页抓取内容、传感器日志等）动态拼接信息，并将其作为上下文输入给 LLM 进行推理。传统提示注入（Prompt Injection）攻击通常依赖于控制主提示的结构或顺序（例如 ...
The Hackers Labs-Securitrona
本文档记录了针对The Hackers Labs平台Securitrona靶机的渗透测试全过程，涉及的漏洞类型：LLM提示词注入、路径遍历、SUID权限滥用
CVE-2026-22785：Orval MCP Code Injection 逃逸导致 RCE 分析
Orval 7.18.0前版本因未转义OpenAPI的summary字段，致MCP生成代码时发生JS注入，引发远程代码执行。
CVE-2026-21877 ：n8n Git 节点远程代码执行漏洞
前言最近n8n的漏洞挺多的，恰好看到https://xz.aliyun.com/news/91090这篇25年的Git 节点 RCE 漏洞分析，一查发现还有个CVE-2026-21877，于是来了兴趣1. 漏洞概述N8N 是一个开源的工作流程自动化平台。在0.121.2及以下版本中，经过认证的攻击者 ...
LibreChat MCP Stdio CVE-2026-22252远程代码执行漏洞分析
LibreChat MCP stdio组件因缺乏输入验证，导致 authenticated 用户可执行任意命令，已在v0.8.2-rc2修复。
AI-Powered Surveillance in Schools
It all sounds pretty dystopian: Inside a white stucco building in Southern California, video cameras compare faces of passersby against a facial recog ...
DevOps & SaaS Downtime: The High (and Hidden) Costs for Cloud-First Businesses
Just a few years ago, the cloud was touted as the “magic pill” for any cyber threat or performance issue. Many were lured by the “always-on” dream, tr ...
New StackWarp Hardware Flaw Breaks AMD SEV-SNP Protections on Zen 1–5 CPUs
A team of academics from the CISPA Helmholtz Center for Information Security in Germany has disclosed the details of a new hardware vulnerability affe ...
CrashFix Chrome Extension Delivers ModeloRAT Using ClickFix-Style Browser Crash Lures
Cybersecurity researchers have disclosed details of an ongoing campaign dubbed KongTuke that used a malicious Google Chrome extension masquerading as ...
Security Bug in StealC Malware Panel Let Researchers Spy on Threat Actor Operations
Cybersecurity researchers have disclosed a cross-site scripting (XSS) vulnerability in the web-based control panel used by operators of the StealC inf ...
AutoVulnPHP：基于大语言模型的两阶段 PHP 漏洞检测与自动定位
作者：Zhiqiang Wang、Yizhong Ding、Zilong Xiao等 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/html/2601.06177v1 摘要 PHP在Web开发中的主导地位受到安全挑战的影响：静态分析缺乏语义深度，导致误报率高；动态分 ...
Weekly Update 487
I thought Scott would cop it first when he posted about what his solar system really cost him last year. "You're so gonna get that stupid AI-slop resp ...
Black Basta Ransomware Leader Added to EU Most Wanted and INTERPOL Red Notice
Ukrainian and German law enforcement authorities have identified two Ukrainians suspected of working for the Russia-linked ransomware-as-a-service (Ra ...
OpenAI to Show Ads in ChatGPT for Logged-In U.S. Adults on Free and Go Plans
OpenAI on Friday said it would start showing ads in ChatGPT to logged-in adult U.S. users in both the free and ChatGPT Go tiers in the coming weeks, a ...
More Problems for Fortinet: Critical FortiSIEM Flaw Exploited
CVE-2025-64155, a command injection vulnerability, was disclosed earlier this week and quickly came under attack from a variety of IP addresses.
GootLoader Malware Uses 500–1,000 Concatenated ZIP Archives to Evade Detection
The JavaScript (aka JScript) malware loader called GootLoader has been observed using a malformed ZIP archive that's designed to sidestep detection ef ...
CISOs Rise to Prominence: Security Leaders Join the Executive Suite
Security professionals are moving up the executive ranks as enterprises face rising regulatory and compliance standards.
AI and the Corporate Capture of Knowledge
More than a decade after Aaron Swartz's death, the United States is still living inside the contradiction that destroyed him. Swartz believed that kno ...
Five Malicious Chrome Extensions Impersonate Workday and NetSuite to Hijack Accounts
Cybersecurity researchers have discovered five new malicious Google Chrome web browser extensions that masquerade as human resources (HR) and enterpri ...
AI System Reduces Attack Reconstruction Time From Weeks to Hours
Pacific Northwest National Labs' expert cybersecurity system, ALOHA, can recreate attacks and test them against organizations' infrastructure to bolst ...
Weekly Update 486
I’m in Oslo! Flighty is telling me I’ve flown in or out of here 43 times since a visit in 2014 set me on a new path professionally and, many years lat ...
KryptoPilot：一种面向自动化密码学漏洞利用的开放世界知识增强型大语言模型智能体
作者：Xiaonan Liu、Zhihao Li、Xiao Lan等 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/html/2601.09129v1 摘要 夺旗赛（CTF）在现代网络安全领域占据核心地位，是培养安全人才、评估源于真实漏洞的攻防技术的主要平台。尽管大 ...
Predator Spyware Sample Indicates 'Vendor-Controlled' C2
Researchers detailed how Intellexa, Predator's owner, uses failed deployments and thwarted infections to strengthen its commercial spyware and generat ...
Winter Olympics Could Share Podium With Cyberattackers
The upcoming Winter Games in the Italian Alps are attracting both hacktivists looking to reach billions of people and state-sponsored cyber-spies targ ...
Vulnerabilities Surge, But Messy Reporting Blurs Picture
MITRE loses its lead as the top reporter of vulnerabilities, while new organizations pump out CVEs and reported bugs in WordPress plug-ins surge.
New Vulnerability in n8n
This isn't good: We discovered a critical vulnerability (CVE-2026-21858, CVSS 10.0) in n8n that enables attackers to take over locally deployed instan ...
Trio of Critical Bugs Spotted in Delta Industrial PLCs
Experts disagree on whether the vulnerabilities in a programmable logic controller from Delta are a five-alarm fire or not much to worry over.
MCP-ITP：一种面向 MCP 的隐式工具投毒自动化框架
作者：Ruiqi Li、Zhiqiang Wang、Yunhao Yao、Xiang-Yang Li 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/html/2601.07395v1 摘要 为了规范基于大型语言模型（LLM）的智能体与其环境之间的交互，模型上下文协议 ...
Retail, Services Industries Under Fire in Oceania
Last year in Australia, New Zealand, and the South Pacific, Main Street businesses like retail and construction suffered more cyberattacks than their ...
Hacking Wheelchairs over Bluetooth
Researchers have demonstrated remotely controlling a wheelchair over Bluetooth. CISA has issued an advisory. CISA said the WHILL wheelchairs did not e ...
Upcoming Speaking Engagements
This is a current list of where and when I am scheduled to speak: I’m speaking at the David R. Cheriton School of Computer Science in Waterloo, Ontari ...
基于大语言模型的特征级对抗攻击：针对安卓恶意软件检测器
作者：Tianwei Lan, Farid Nait-Abdesselam 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/html/2512.21404v1 摘要 安卓恶意软件在规模和复杂性上的快速增长，推动了机器学习（ML）技术在可扩展、高精度恶意软件检测中的广泛 ...
Patch Tuesday, January 2026 Edition
Microsoft today issued patches to plug at least 113 security holes in its various Windows operating systems and supported software. Eight of the vulne ...
1980s Hacker Manifesto
Forty years ago, The Mentor—Loyd Blankenship—published "The Conscience of a Hacker" in Phrack. You bet your ass we're all alike… we've been spoon-fed ...
Who Decides Who Doesn’t Deserve Privacy?
Remember the Ashley Madison data breach? That was now more than a decade ago, yet it arguably remains the single most noteworthy data breach of all ti ...
模仿游戏：利用大型语言模型作为聊天机器人打击基于聊天的网络犯罪
作者：Yifan Yao, Baojuan Wang, Jinhao Duan等 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/html/2512.21371v1 摘要 基于聊天的网络犯罪已成为一种普遍存在的威胁，攻击者利用实时消息平台实施诈骗，这类诈骗依赖于建立信 ...
Corrupting LLMs Through Weird Generalizations
Fascinating research: Weird Generalization and Inductive Backdoors: New Ways to Corrupt LLMs. Abstract LLMs are useful because they generalize so well ...
DEVCORE 的 Synology BeeStation RCE 与一种新型 SQLite 注入 RCE 技术（CVE-2024-50629~50631）
作者：Kiddo 原文链接：https://kiddo-pwn.github.io/blog/2025-11-30/writing-sync-popping-cron 引言 9 月，在为 Pwn2Own Ireland 2025 做准备时，我翻遍了 Synology NAS 的 N-day 漏洞寻找 ...
Friday Squid Blogging: The Chinese Squid-Fishing Fleet off the Argentine Coast
The latest article on this topic. As usual, you can also use this squid post to talk about the security stories in the news that I haven't covered. Bl ...
DeepSeek-R1：通过强化学习激发大型语言模型的推理能力
作者：DeepSeek-AI 译者：知道创宇404实验室翻译组 原文链接：https://arxiv.org/pdf/2501.12948 摘要 通用推理是人工智能领域一项长期且艰巨的挑战。近年来，以大型语言模型（LLMs）（Brown 等人，2020；OpenAI，2023）和思维链提示（Wei ...
Who Benefited from the Aisuru and Kimwolf Botnets?
Our first story of 2026 revealed how a destructive new botnet called Kimwolf rapidly grew to infect more than two million devices by mass-compromising ...
V8 安全简单记录
作者：陈广 原文链接：https://blog.chain0x0.com/blog/%E6%B5%8F%E8%A7%88%E5%99%A8%E5%AE%89%E5%85%A8%E7%AE%80%E5%8D%95%E8%AE%B0%E5%BD%95 序言 在src漏洞挖掘中总是遇到高版本chrome浏 ...
Weekly Update 485
15 mins and 40 seconds. That's how long it took to troubleshoot the first tech problem of 2026, and that's how far you'll need to skip through this vi ...
ComfyUI-Manager 远程代码执行风险通告
近期腾讯玄武实验室发现可视化 AI 工作流工具 ComfyUI 的官方扩展组件 ComfyUI-Manager 中存在一个高危漏洞（CVE-2025-67303）。利用该漏洞可在无需任何账号的情况下远程入侵安装 ComfyUI 的系统。玄武实验室在发现漏洞后向 ComfyUI 官方进行了报告，目前该 ...
The Kimwolf Botnet is Stalking Your Local Network
The story you are reading is a series of scoops nestled inside a far more urgent Internet-wide security advisory. The vulnerability at issue has been ...
Happy 16th Birthday, KrebsOnSecurity.com!
KrebsOnSecurity.com celebrates its 16th anniversary today! A huge "thank you" to all of our readers -- newcomers, long-timers and drive-by critics ali ...
Weekly Update 484
I think the start of this week's video really nailed it for the techies amongst us: shit doesn't work, you change something random and now shit works ...
Weekly Update 483
Building out an IoT environment is a little like the old Maslow's Hierarchy of Needs. All the stuff on the top is only any good if all the stuff on th ...
Dismantling Defenses: Trump 2.0 Cyber Year in Review
The Trump administration has pursued a staggering range of policy pivots this past year that threaten to weaken the nation’s ability and willingness t ...
Weekly Update 482
Perhaps it's just the time of year where we all start to wind down a bit, or maybe I'm just tired after another massive 12 months, but this week's vid ...
Most Parked Domains Now Serving Malicious Content
Direct navigation -- the act of visiting a website by manually typing a domain name in a web browser -- has never been riskier: A new study finds the ...
Processing 630 Million More Pwned Passwords, Courtesy of the FBI
The sheer scope of cybercrime can be hard to fathom, even when you live and breathe it every day. It's not just the volume of data, but also the exten ...
Microsoft Patch Tuesday, December 2025 Edition
Microsoft today pushed updates to fix at least 56 security flaws in its Windows operating systems and supported software. This final Patch Tuesday of ...
Drones to Diplomas: How Russia’s Largest Private University is Linked to a $25M Essay Mill
A sprawling academic cheating network turbocharged by Google Ads that has generated nearly $25 million in revenue has curious connections to a Kremlin ...
量子计算机距离攻破 RSA-2048 还有多远
Author: Guancheng Li of Tencent Xuanwu Lab 在当今数字世界中，RSA‑2048 与 ECC 等经典公钥密码是最广泛应用的加密标准，支撑着网络安全、金融交易和隐私保护的底层信任。然而，这一基石正面临量子计算的潜在威胁。理论上，量子计算机能够以远快于经典计算机的 ...
我们的AI发现了一个零知识证明库的漏洞，Sam Altman的项目也用了这个库
Author: Guancheng Li, Xiaolin Zhang and Yang Yu of Tencent Xuanwu Lab 2025 年 8 月，腾讯玄武实验室的阿图因自动化漏洞挖掘引擎在零知识证明库 gnark 中发现了一个高危漏洞（CVE-2025-57801，CVSS 8.6） ...
玄武在BlackHat揭示劫持智能体达成RCE的新方法
Author: Jiashuo Liang and Guancheng Li of Tencent Xuanwu Lab 0x00 前言大语言模型（LLM）正在从简单的对话工具演化为能够编写代码、操作浏览器、执行系统命令的智能体。随着大模型应用的演进，提示词注入攻击的威胁也在不断升级。 设想这样一个 ...
腾讯和清华的最新研究：利用“长度侧信道”绕过5G/4G/WiFi网络的加密
Author: Guancheng Li of Tencent Security Xuanwu Lab 在当今数字化时代，5G、4G和Wi-Fi等无线通信技术已成为我们日常生活的重要基础设施。这些网络普遍采用先进的加密协议，理论上能够有效保护用户通信安全。然而，近期由我们腾讯玄武实验室与清华大学陈建 ...