ARGUS: Defending LLM Agents Against Context-Aware Prompt Injection
随着大型语言模型（LLM）代理的兴起，其通过集成工具调用、技能和外部知识，显著提升了自动化能力，但也引入了新的安全风险。其中，提示注入攻击（Prompt Injection）已成为主要威胁：攻击者将恶意指令嵌入代理工作流中，劫持代理行为。然
2026-05-06 归档
今日共收录 1285 条安全情报，包含 277 个 CVE，391 篇安全通告，以及 322 篇研究论文。
← 返回归档列表MAGE: Safeguarding LLM Agents against Long-Horizon Threats via Shadow Memory
随着大语言模型驱动的智能体（LLM agents）被部署到复杂的现实世界任务中，它们面临一类日益增长的攻击：利用扩展的用户-智能体-环境交互，在单轮对话中难以实现的恶意目标。这类“长程威胁”对关键领域智能体的安全部署构成重大风险。本文提出M
Dependency-Aware Privacy for Multi-turn Agents
本文研究多轮交互场景下LLM智能体的隐私保护问题。现有基于度量差分隐私的提示清洗器将每次服务调用视为独立发布，但攻击者可通过跨轮联合观测恢复隐私属性，导致隐私随发布次数增加而衰减。作者指出这种退化是根本性的：当隐私属性是计算图的根节点时，对
Evaluating Retrieval-Augmented Generation for Explainable Malware Analysis
该论文针对大型语言模型（LLM）在恶意软件分析中的应用进行了实证研究，重点评估检索增强生成（RAG）技术对解释质量的影响。研究背景是：安全分析师常借助LLM来自动总结和解释恶意软件行为，而RAG被认为可以通过注入外部安全知识来提升解释质量。
EvoPoC: Automated Exploit Synthesis for DeFi Smart Contracts via Hierarchical Knowledge Graphs
本文提出了一个名为EvoPoC的知识驱动智能体系统，用于端到端的DeFi智能合约漏洞检测与利用合成。核心思路是将利用合成视为结构化推理问题，而非代码生成任务，因此需要协议语义、失败根因和利用原语的知识。EvoPoC将知识组织为层次知识图谱（
jdbi3-freemarker Vulnerable to Improper Neutralization of Special Elements Used in FreeMarker Template Engine
# Summary **Description** An Improper Neutralization of Special Elements Used in a Template Engine (CWE-1336) vulnerab
Grav is Vulnerable to XXE via SVG Upload
Dear Grav Security Team, A security vulnerability was discovered in Grav CMS that allows authenticated attackers to rea
@evomap/evolver's validator sandbox allowlist permits `npm`/`npx`, yielding RCE from Hub-delivered validation tasks via lifecycle scripts
## Summary The validator-mode sandbox executor (`src/gep/validator/sandboxExecutor.js`) places `npm` and `npx` in its h
CrowdStrike Named a Leader in Frost & Sullivan 2026 Radar for Cloud-Native Application Protection Platforms
CrowdStrike Named a Leader in Frost & Sullivan 2026 Radar for Cloud-Native Application Protection Platforms
CrowdStrike Expands Real-Time Cloud Detection and Response to Google Cloud
CrowdStrike Expands Real-Time Cloud Detection and Response to Google Cloud
CrowdStrike Falcon Cloud Security Delivered 264% ROI Through Unified Cloud Protection
CrowdStrike Falcon Cloud Security Delivered 264% ROI Through Unified Cloud Protection
Security & Identity
Security & Identity
SAP on Google Cloud
SAP on Google Cloud
Inside Google Cloud
Inside Google Cloud
Google Cloud Next & Events
Google Cloud Next & Events
Google Cloud Consulting
Google Cloud Consulting
Transform with Google Cloud
Transform with Google Cloud
Threat IntelligenceNorth Korea-Nexus Threat Actor Compromises Widely Used Axios NPM Package in Supply Chain AttackBy Google Threat Intelligence Group • 16-minute read
Threat IntelligenceNorth Korea-Nexus Threat Actor Compromises Widely Used Axios NPM Package in Supply Chain AttackBy Goo
Threat IntelligenceRansomware Under Pressure: Tactics, Techniques, and Procedures in a Shifting Threat LandscapeBy Google Threat Intelligence Group • 53-minute read
Threat IntelligenceRansomware Under Pressure: Tactics, Techniques, and Procedures in a Shifting Threat LandscapeBy Googl
Google Cloud Products
Google Cloud Products