LayerX found a flaw in Anthropic’s Claude Chrome extension that lets other extensions inject hidden instructions and take control of the AI agent. The issue enabled unauthorized access to Google Drive, Gmail, and GitHub data, and LayerX says the partial fix still left some takeover paths open. #Anthropic #Claude #LayerX #Chrome #GoogleDrive #Gmail #GitHub
Keypoints
- LayerX discovered a Chrome extension flaw in Anthropic’s Claude AI.
- Any other browser extension could inject hidden instructions without special permissions.
- The bug allowed prompt injection, guardrail bypass, and cross-site actions.
- Proof-of-concept attacks exposed Google Drive files, Gmail activity, and GitHub source code.
- LayerX said Anthropic’s May 6 fix was only partial and some takeover scenarios still worked.
Read More: https://cyberscoop.com/claude-chrome-extension-allows-plugins-to-hijack-ai/