GitGuardian detects 24,008 secrets leaked in public AI agent configuration files
Tool · Dev.to · stat: 3.2% leak GitGuardian reports that AI coding assistants including Cursor, Claude Code, and GitHub Copilot are exposing credentials across local environments. The security firm's…
Tool · Dev.to · stat: 3.2% leak
GitGuardian reports that AI coding assistants including Cursor, Claude Code, and GitHub Copilot are exposing credentials across local environments. The security firm's 2026 study identifies 24,008 unique secrets in public Model Context Protocol configuration files, alongside a 3.2% credential leak rate in commits assisted by Claude Code.
AI coding agents bypass standard repository scanners, creating quiet security blindspots. Startups using Cursor or Claude Code must audit local configuration files immediately to prevent API keys from leaking into public repositories.
Every claim ties to a primary source. See our methodology.