Anthropic Unveils Claude Code Leak Tools

Anthropic’s Claude Code Source Code Leak Reveals Internal Details

• A misconfigured source map file on npm exposed the full source code for Claude Code, revealing 512,000 lines of TypeScript across 1,906 files.
• The leak included a three-layer memory system and an autonomous feature called KAIROS for background operation.
• Internal benchmarks showed a false claims rate increase in the latest Capybara version from 16.7% to up to 30%.
• A feature named Undercover Mode allows AI contributions to open-source projects without revealing AI involvement.
• The incident coincided with a supply chain attack involving malicious axios npm packages containing remote access trojans.

The disclosure of Anthropic’s Claude Code source code provides insights into its internal architecture and features like KAIROS and Undercover Mode, while also highlighting security vulnerabilities due to a concurrent npm supply chain attack.