Anthropic's Claude Code Source Leaked
The source code for Anthropic's AI assistant Claude was accidentally leaked due to a misconfigured build script, revealing details about its internal architecture and unreleased models.
Why it matters
This leak underscores the need for robust security measures in AI development, as even small oversights can expose sensitive technology.
Key Points
- 1Anthropic's Claude code source was found in a public npm package due to a build script error
- 2The leaked code revealed details about Claude's self-healing memory, unreleased model codenames, and built-in agent swarms
- 3The leak also showed Anthropic's logic for contributing to open-source repos without explicit AI attribution
Details
Security researcher Chaofan Shou discovered that the source code for Anthropic's AI assistant Claude was accidentally included in a public npm package. The 57MB source map file provided a complete decoder back to Anthropic's original TypeScript code, exposing details about Claude's internal architecture. The leaked code revealed features like a three-layer self-healing memory system, codenames for unreleased models, and the ability for Claude to autonomously spawn parallel sub-agents. It also showed Anthropic's approach to contributing to open-source projects without explicit AI attribution. While Anthropic has stated that no model weights or customer data were compromised, the leak has made Anthropic's proprietary technology publicly accessible, highlighting the importance of secure software development practices even for leading AI companies.
No comments yet
Be the first to comment