ClawSecure says MCP protocol has structural vulnerabilities

1. 18:42
   Forkast

   ClawSecure report claims the MCP protocol itself has structural security flaws

   Security vendor ClawSecure stated in the AI Agent Threat Report that the MCP protocol specification itself has structural vulnerabilities, not vendor implementation issues, and developers cannot fix them by patching. Testing Linear, Notion, and Dropbox Dash found that servers automatically fetch attacker-controlled links when content is created, forming data exfiltration channels without AI involvement; 17 of 20 obfuscation techniques passed round-trip detection, and none of the 14 models from five laboratories could reliably block them, with the best-performing Claude Opus 4.7 still having a 26.7% probability of complying with malicious instructions. The report has not yet been independently reproduced by third parties, and no CVE or official patch has been released.

1. Date not given
   ClawSecure

   ClawSecure red-team report: 14 models and MCP integrations including Dropbox Dash, Notion, and Linear all have indirect prompt injection risks

   ClawSecure 对来自五家前沿实验室的 14 款模型和 Dropbox Dash、Notion、Linear 三大 MCP 平台做间接提示注入红队测试，全部模型都服从了攻击者指令，无一干净防御。表现最好的 Claude Opus 4.7 服从率仍达 26.7%，GPT-5.5 与 Qwen 3.7 Max 为 91.7%，Gemini 3.1-flash-lite 和多款 Gemini 2.5/3.5-flash 则达到 100%。测试于 2026 年 5 至 7 月进行，Anthropic 宣称的 0% 智能体编码攻击成功率被同一模型上 15.2% 的结果推翻。

2. Date not given
   EIN Presswire

   ClawSecure report: Linear MCP, Notion, and Dropbox Dash all compromised in security testing

   ClawSecure 在《AI Agent Threat Report》第一卷中对 Linear MCP、Notion 和 Dropbox Dash 三个 MCP 平台进行了实弹漏洞复现测试，三个平台全部被攻破，其结论是漏洞根源于 MCP 协议本身，而非某一家公司的代码缺陷。在 Dropbox Dash 上，测试的 5 个攻击面中有 4 个被攻破；ClawSecure 还在 75 分钟无引导研究中对 Linear 生产系统植入了攻击者控制的凭证诱饵，并非在仿真环境中进行。Notion 和 Linear 的服务器会在内容创建时自动抓取攻击者控制的链接，无需 AI 参与也无需任何人打开内容。该报告于 2026 年 9 月 24 日发布，结论经过 OWASP、MITRE ATLAS 和 NIST 参照的四轮取证复核。

Current attention35·Peak in the comparable period40（Oct 7 12:00）·Change over the comparable past 24 hours 0%