Claude Code leaks secrets into 1,122 GitHub files
Original: Claude Code's "Don't Ask Again" exposes passwords in 1,122 public GitHub files
Why This Matters
Agentic coding tools are now a credible new vector for secrets sprawl at scale.
Apiiro found Claude Code's 'Don't Ask Again' feature silently writes credentials and API keys into plain-text config files, exposing secrets in 1,122 public GitHub repositories discovered via search.
Security firm Apiiro has identified a significant secrets-leakage pattern tied to Claude Code, Anthropic's agentic coding tool. The culprit is the 'Don't Ask Again' permission feature: when developers suppress repeated permission prompts, Claude Code stores those decisions—along with associated credentials and API keys—in plain-text configuration files. Those files then get committed to version control and pushed to public GitHub repositories.
Apiiro's researchers found 1,122 public GitHub files containing secrets exposed through this mechanism. The issue is behavioral rather than a traditional software vulnerability: developers opt out of friction, the tool silently persists their choices to disk, and standard git workflows do the rest. The discovery highlights a broader pattern emerging with agentic AI dev tools—convenience features that reduce interruptions can quietly expand the attack surface by writing sensitive data to files that weren't historically secret.
Apiiro positions itself as an application security platform focused on agentic development risks, so the firm has a commercial interest in surfacing this category of finding. That said, the 1,122-file count is a concrete, verifiable data point from GitHub's public index.