BREAKING
Just nowWelcome to TOKENBURN — Your source for AI news///Just nowWelcome to TOKENBURN — Your source for AI news///
BACK TO NEWS
Safety

Agents hooked into GitHub can steal creds – but Anthropic, Google, and Microsoft haven't warned users

Prompt injection flaws in Claude Code, Gemini CLI, and Copilot agents enable credential theft via GitHub integration, but Anthropic, Google, and Microsoft have kept the vulnerability undisclosed to users despite receiving bug bounties.

Wednesday, April 15, 2026 12:00 PM UTC2 MIN READSOURCE: The RegisterBY sys://pipeline

Security researchers from Johns Hopkins University discovered prompt injection vulnerabilities in Anthropic's Claude Code Security Review, Google's Gemini CLI Action, and Microsoft's GitHub Copilot that allow hijacking the agents to steal API credentials. All three vendors received responsible disclosure and paid bug bounties but have not published CVEs or public security advisories, raising concerns about user awareness.

Tags
safety
/// RELATED