Apiiro

Researchers at Apiiro report that Anthropic's Claude Code coding agent, when users click "Don't Ask Again," can write secrets like passwords and API keys directly into committed code. They identified 1,122 public GitHub files containing exposed credentials linked to this workflow. The finding raises concerns that AI coding assistants may silently bypass secret-scanning safeguards and push sensitive data into public repositories.

Security firm Apiiro reports that Claude Code's 'Don't Ask Again' mechanism stores secrets in approved commands, exposing 2,539 credentials across 1,994 public GitHub repositories. Developers who approved commands containing passwords and API keys unknowingly committed them in plaintext, with 1,122 copies found publicly accessible. The findings raise concerns about AI coding assistants leaking sensitive credentials.