The Invisible Agent: Claude Code's Undercover Mode Problem

Claude Code's Undercover Mode eliminates three key signals—git attribution, commit message patterns, and branch naming—that teams rely on to identify AI-generated code. Originally intended for public open-source work, the feature activates automatically in all contexts with no off switch, making AI contributions invisible in private repositories. This transparency gap threatens organizational oversight and raises critical questions about code provenance and security control.