Claude Code Source Code Leaked After Anthropic Packaging Mistake

Anthropic has acknowledged that a recent release of its Claude Code CLI unintentionally exposed internal source code, after a packaging error made the full codebase publicly accessible.

The issue occurred when version 2.1.88 of the Claude Code npm package was published with a source map file included. That file allowed access to nearly 2,000 TypeScript files and more than 512,000 lines of code, which were quickly shared online and mirrored in public repositories.

Security researcher Chaofan Shou first identified the exposure, and the code was soon uploaded to GitHub, where it was widely copied and analyzed by developers.

Anthropic said the incident was not the result of a breach and did not involve user data. “Earlier today, a Claude Code release included some internal source code. No sensitive customer data or credentials were involved or exposed. 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 company said.

The leak has given developers and competitors an unusually detailed look into how Claude Code is built. Early analysis has focused on areas like memory handling systems and internal tooling architecture, with some developers noting the scale and complexity of the codebase.

While the company’s core AI models were not exposed, the incident reveals internal implementation details that could be used to inform competing products or identify potential vulnerabilities. Developers have already begun dissecting components such as the system used to manage memory and validate stored information.

The code release also raises concerns about how quickly developer tools in the AI space are being shipped and distributed. Claude Code is part of a rapidly evolving category of AI-powered developer tools, where companies are racing to expand capabilities and adoption.

Because the tool integrates closely with developer workflows, the exposure provides insight into both its architecture and how it operates in production environments. It also creates potential risks if bad actors use the information to probe for weaknesses in safeguards.

Despite that, the long-term impact remains unclear, as the category continues to move quickly and new iterations of tools are released at a rapid pace.

This analysis is based on reporting from Ars Technica.

Image courtesy of Samuel Axon.

This article was generated with AI assistance and reviewed for accuracy and quality.