SPA: Securing Persistent LLM Agents Across Queries with Plan-First Information-Flow Control
本文针对大型语言模型（LLM）智能体在跨查询持久化场景下的安全威胁，提出了一种名为 SPA 的“计划优先”架构。研究背景是：现有防御通常只保护规划阶段或单个工具交互，但持久化智能体面临更广泛的攻击面——攻击者控制的数据可能改变控制流、进入安
2026-08-31 归档
今日共收录 884 条安全情报，包含 132 个 CVE，408 篇安全通告，以及 86 篇研究论文。
← 返回归档列表LLM-Based Agents for Software and Systems Security: Approaches, Applications, and Assessment
本文是一篇针对大语言模型（LLM）智能体在软件与系统安全领域应用的系统性文献综述，覆盖2023至2026年间经过同行评审的相关研究。论文指出，安全运维工作通常是流程化的：分析师需要检查异构工件、形成假设、调用工具、解读输出并修正计划，而基于
BEACON: Behavior-Anchored Cross-Source Knowledge Graph Construction for Cyber Threat Intelligence
本文提出 BEACON，一个基于行为锚点的跨来源威胁情报知识图谱构建框架，用于从非结构化的网络威胁情报（CTI）报告中自动提取并整合知识。现有方法主要处理单篇报告，且提取信息不完整，忽略了同一威胁在跨来源报告中可能使用不同名称的问题。作者的
PLCBench: Can Autonomous LLM Agents Turn PLC Access into Sustained Physical Impact?
PLCBench 论文提出并实验验证了一个面向真实工业控制系统（ICS）的硬件在环（HIL）安全评估框架，重点研究自主大语言模型（LLM）代理能否将网络可达的可编程逻辑控制器（PLC）访问转化为持续的物理影响。现有评估大多聚焦于数字任务或
Quantization-Triggered Backdoors in Language Models: Cross-Quantizer Transferability and the Validation--Deployment Gap
本文针对大语言模型（LLM）后训练量化这一常见优化流程，提出并验证了一种新型后门攻击范式。量化通常被视为语义中立，但在实际工作流中，模型通常先在源精度（如FP16）上完成评估，之后再应用量化并直接部署，这种'验证-部署差距'（validat
A vulnerability was found in TOTOLINK NR1800X 9.1.0u.6681_B20230703. Impacted is the function setUploadSetting of the file /cgi-bin/cstecgi.cgi. The manipulation of the argument...
CVE-2026-82616
该漏洞存在于 TOTOLINK NR1800X 路由器（固件版本 9.1.0u.6681_B20230703）的 /cgi-bin/cstecgi.cgi 接口中。具体函数为 setUploadSetting，该函数未对用户输入的 File
Cloud & Application Security
Cloud & Application Security
From Scanner to Stealer: Inside the trivy-action Supply Chain Compromise
From Scanner to Stealer: Inside the trivy-action Supply Chain Compromise
Next-Gen Identity Security
Next-Gen Identity Security
CrowdStrike Expands Identity Leadership with OpenID and IDPro
CrowdStrike Expands Identity Leadership with OpenID and IDPro
cloud architecture assessment
cloud architecture assessment
ransomware and multifaceted extortion defense
ransomware and multifaceted extortion defense
Google Cloud Next 2026Catch-up sessions on the keynotes and select sessions are now available on demand.Explore content on-demand
Google Cloud Next 2026Catch-up sessions on the keynotes and select sessions are now available on demand.Explore content
The Identity Problem Hiding in AI Agent Deployments
The Identity Problem Hiding in AI Agent Deployments
94% of Organizations Report Cloud Breaches: CrowdStrike State of CDR Survey
94% of Organizations Report Cloud Breaches: CrowdStrike State of CDR Survey
Falcon Cloud Security June 2026 Release: Updates for Azure and Google Cloud
Falcon Cloud Security June 2026 Release: Updates for Azure and Google Cloud
Falcon Cloud Security July 2026 Release: Helping Security Teams Move Faster in the Cloud
Falcon Cloud Security July 2026 Release: Helping Security Teams Move Faster in the Cloud
CrowdStrike Named Strongest Overall Leader in 2026 Frost Radar™: Cloud Workload Protection Platforms
CrowdStrike Named Strongest Overall Leader in 2026 Frost Radar™: Cloud Workload Protection Platforms
TerminalFix Uses Fake Cloudflare CAPTCHAs to Deploy Reverse-Tunnel Backdoor
Microsoft has disclosed details of a new ClickFix variant, dubbed TerminalFix, that aims to trick users into running a m
RAIFLE: Reconstruction Attacks on Interaction-based Federated Learning with Adversarial Data Manipulation.
本文提出并研究了针对交互式联邦学习（Interaction-based Federated Learning）的隐私重构攻击方法，命名为RAIFLE。联邦学习旨在通过不共享原始数据、仅交换模型参数或梯度来保护用户隐私，但近年来已有研究表明，