CrowdStrike SafeMind: When the Best Offense Builds the Best DefenseSep 17, 2026
CrowdStrike SafeMind: When the Best Offense Builds the Best DefenseSep 17, 2026
2026-09-19 归档
今日共收录 1784 条安全情报，包含 814 个 CVE，581 篇安全通告，以及 100 篇研究论文。
← 返回归档列表‘Nudify’ apps: What to do if someone makes a fake nude of you
Whether you’re a victim, the parent of a victim, or just concerned, here’s what you can do about fake nude images
Gyazo server flaw exploited to steal 23.6 million user records
The Gyazo image-sharing platform has confirmed it suffered a data breach after hackers exploited a server vulnerability
Fake LastPass Authenticator GitHub repos push new Rapuncel infostealer
An ongoing malware campaign uses SEO-optimized GitHub repositories to impersonate well-known software firms to push a pr
Secure enterprise sharing with access reviews for Microsoft 365
Microsoft 365 makes sharing files easy, but access can remain long after its original purpose has ended, leaving organiz
Webinar: Which Google Workspace security controls actually matter?
Fast-growing companies face countless recommendations for securing Google Workspace, but not every control provides the
Microsoft fixes bug behind ‘Defender Antivirus is turned off’ alerts
Microsoft has resolved a known issue that causes incorrect alerts warning that Defender Antivirus was turned off after i
New Check Point flaw lets hackers execute code with root privileges
Check Point Software has released security updates to address a critical vulnerability that can let attackers execute co
Microsoft fixes broken copy and paste for Excel 2016 users
Microsoft has fixed a known issue that causes copy-and-paste failures for some Excel users after installing the Septembe
Transparent Tribe Deploys New Rust Backdoor Using Private GitHub Repositories for C2
The Pakistan-aligned threat group tracked as Transparent Tribe (aka APT36 and Earth Karkaddan) has been attributed to a
WeaselBiscuit Stealer Spreads via 13 npm Packages to Harvest Chrome Extension Storage
Cybersecurity researchers have discovered a cluster of 13 npm packages that have been found to deliver a previously undo
In Other News: Ransomware Developer Sentenced, Plugin4Shell AI Attack, Critical SAP Flaw
Noteworthy stories that might have slipped under the radar: Mandiant's 2026 AI risk report, PhantomRaven malware used by
Microsoft Patches 18 Vulnerabilities in AI, Cloud Products
Microsoft fixed vulnerabilities across Azure and AI-branded products, with privilege escalation flaws accounting for the
Cybersecurity in Power Grids: Standards and Research Challenges
本文是一篇面向智能电网（Smart Grid）网络安全的综述性研究论文，作者来自加泰罗尼亚地区高校与研究机构，投稿分类为 eess.SY（系统工程）。文章的核心出发点在于强调信息技术（IT）与运营技术（OT）环境之间的关键差异：IT 侧关注
Measuring and Exploiting Implicit Trust in LLM Tool-Calling Pipelines
该论文聚焦 Model Context Protocol (MCP) 使能的大模型工具调用链路中的「隐式信任」问题。MCP 允许 LLM 调用外部工具，但每次工具交互都会通过多个输入通道（工具描述、工具执行结果、采样消息 sampling
CASHEWS: Source Preprocessor for LLM-based Malicious Package Detection
背景与问题:近年来,恶意 npm 包检测工具开始借助 LLM 对源代码的语义理解能力,在大规模场景下识别包的恶意意图,并在 Shai-Hulud 等近期供应链攻击事件中体现出实用价值。然而,LLM 的上下文窗口有限,攻击者正是利用这一点进行
The More It Says, the More You Pay: A Black-Box Audit of Provider-Side Token Inflation in LLM Services
在按 token 计费的 LLM 服务中，输出越长用户付费越多，这为不诚信的服务提供方创造了经济动机：在不明显损害任务可用性的前提下，暗中操纵生成过程以膨胀输出 token 数量。论文将此类行为定义为“提供方侧 Token 膨胀攻击”（Pr
The Illusion of Local Privacy: Confidentiality Boundary Failures in Consumer LLM Serving Systems
该论文研究消费级本地大模型（local LLM）部署中的提示词机密性问题。业界普遍认为在本机运行推理比云端更私密，因为提示词不离开设备；作者对此提出质疑，并论证「本地推理」本身并不足以保证提示词保密，还取决于服务软件在推理前、推理中和推理后
ALIBI: Adversarial Legitimacy Injection in Binary Input against LLM Malware Analyzers
大型语言模型正被集成到恶意软件分诊流程中, 作为推理组件汇总静态证据并给出面向分析师的判定结论。本文指出, 这种推理能力本身会引入新的攻击面, 并提出 ALIBI——一种针对前沿 LLM 恶意软件分析器的“语义掩护故事”攻击。ALIBI 不
Closed-World Resolution Against Tool Hallucination in LLM Agents
该论文是一篇以测量与基准为核心的实证研究，聚焦工具增强型 LLM Agent 的一个结构性盲区：模型会调用根本不存在的工具，或传入任何 schema 都未声明的参数，即"工具幻觉"。作者指出，现有的两类防御——工具选择（从候选中挑对工具）与