Anthropic, an AI safety and research company, confirmed that a portion of the internal source code for Claude Code, its AI-powered developer tool, was accidentally released due to a packaging error. The incident exposed approximately 2,000 files and 500,000 lines of code.
The leak included internal architecture and instructions for AI coding agents, but Anthropic stated no customer data or credentials were compromised. The code was briefly mirrored on GitHub, a cloud-based platform that acts as a central hub for developers to store, manage, share, and collaborate on code projects, and it was widely shared on social media before Anthropic issued copyright takedown requests. Developers analyzing the files reportedly found blueprints for an “always-on” AI agent and other internal prototypes.
While the underlying Claude AI models remain confidential, industry analysts suggest the exposed logic could provide competitors with insights into Anthropic’s agentic AI frameworks.
This marks the second data-related incident for Anthropic in recent weeks, following reports of internal files being stored on publicly accessible systems. The leak occurs at a time of significant growth for the company, with paid subscriptions doubling this year.
