Overview
- Claude Code’s v2.1.88 update on Tuesday shipped with a 60MB source‑map file that revealed roughly 1,900 TypeScript files, a mistake first flagged by security researcher Chaofan Shou before the code was mirrored across GitHub and X posts with tens of millions of views.
- Anthropic said the incident stemmed from human error in release packaging and confirmed no customer data or model weights were exposed, then pulled the npm version and pursued DMCA takedowns against reposts.
- Security firms warned that the leak enables tailored exploits because attackers can now study Claude Code’s data flow, with researchers also flagging typosquatted npm packages and a brief Axios supply‑chain compromise that led to advice to downgrade affected installs and rotate secrets.
- Developers poring over the files reported internal designs and unreleased features such as a three‑layer memory system, a persistent agent called KAIROS, a background “dream” mode, and an Undercover Mode for low‑profile open‑source contributions, though the core AI model weights were not included.
- Claude Code creator Boris Cherny later said a missed manual deploy step caused the error and that the team is moving to add more automation, in what is the second Anthropic exposure in a week that has intensified scrutiny of its release controls.