Anthropic Claude Code Leak Exposes Architecture, Unshipped Features

San Francisco: Anthropic has accidentally exposed the source code of Claude Code, its flagship AI coding tool, through a packaging error in a routine npm update. Version 2.1.88 of the Claude Code package was published to the public npm registry with a 59.8MB JavaScript source map file attached, a debugging artifact that enabled anyone to reconstruct nearly 2,000 TypeScript files and over 500,000 lines of code.

Security researcher Chaofan Shou flagged the exposure on X at 4:23am ET on March 31. Within hours, the codebase was mirrored to GitHub, where it surpassed 84,000 stars and 82,000 forks. Anthropic pulled the affected version the same day.

In a statement confirmed to multiple outlets, Anthropic said no customer data or credentials were involved. “This was a release packaging issue caused by human error, not a security breach. We’re rolling out measures to prevent this from happening again.” The exposed code relates to Claude Code agentic harness, not the underlying AI model itself.

What the leak revealed goes beyond source code. Developers digging through the files found 44 feature flags for capabilities fully built but not yet shipped, including a persistent background agent mode, session memory consolidation, and remote control from a phone or browser. The leak also confirmed internal references to an upcoming model, separately identified as Capybara, that Anthropic is actively preparing to launch.

The timing is damaging. Anthropic is preparing for an IPO expected later this year. Claude Code generates $2.5 billion in annualised revenue and is central to the company’s enterprise growth story. Competitors now have a detailed engineering roadmap for how Anthropic builds production-grade agentic systems.

This is the second major exposure incident in weeks. Days earlier, Fortune reported that Anthropic had left nearly 3,000 internal files publicly accessible, including a draft blog post detailing the upcoming Capybara model.