Scope: what this does not show
Specific products and versions, most since patched.
Corroborated: Supported by at least two independent sources.
Evidence
Apr 1, 2025
Aug 5, 2025
Aug 12, 2025
Feb 25, 2026
Check Point: Claude Code project files could run commands and leak API keys before trust prompt
Related rather than direct: repository configuration ran at or before the trust prompt, not after a later change.
Oct 13, 2026
Researcher reports Copilot in Visual Studio and SSMS could rewrite its own settings and reach code execution (CVE-2026-103572)
The agent itself could rewrite its protected tool configuration through a path-check bypass, per the researcher.
Key questions that rely on this finding
- Where are deployed AI agents actually being exploited?Exploitation often reaches deployed agents through connectors, credentials, tools and packages.