Overview
- Anthropic’s @anthropic-ai/claude-code package, updated Tuesday as version 2.1.88, included a 59.8 MB source map—a debug file that links built code back to original sources—that let researchers reconstruct roughly 1,900 TypeScript files totaling about 512,000 lines.
- Security researcher Chaofan Shou flagged the file on X, and copies spread to GitHub within hours as posts about the code drew tens of millions of views.
- Anthropic said the release was a human packaging error, removed the version, and said no customer data, credentials, or model weights were exposed.
- Analysts reviewing the files described internal architectures and unreleased features, including a three-layer self-healing memory system, a KAIROS background agent, and an “Undercover Mode” for discreet open-source contributions.
- Security firms warned of targeted exploits and supply-chain risks, pointing to typosquatted npm packages and a separate axios incident during the March 31 UTC window, and advised users to downgrade or use the native installer, audit dependencies, and rotate any keys; the leak is the company’s second exposure reported in a week, heightening scrutiny of its release controls.