Anthropic accidentally disclosed 512,000 lines of proprietary source code for its artificial intelligence coding tool Claude Code. The incident also raised concerns over the company's planned initial public offering targeted for the fourth quarter this year.

BeInCrypto reported on Tuesday that the leak occurred when debug files were included in a regular npm update on March 31.

Security researcher Chaofan Shou (차오판 쇼우) identified an exposed source map file in Claude Code version 2.1.88 and posted a download link on X. The codebase spread to GitHub within hours and drew tens of thousands of forks before Anthropic followed with DMCA takedown requests.

The industry focus is that the leak may be difficult to contain as a one-off incident. The outlet said rapid copying and redistribution has made it effectively hard to reverse. Canadian developer of Korean descent Sigrid Jin (시그리드 진) started a "clean room" Python rewrite project based on the leaked version, and his repository, claw-code, was reported to have logged 50,000 GitHub stars shortly after it was made public.

Anthropic confirmed the leak to multiple media outlets and attributed it to a packaging error caused by "human error." The incident was the second one after a separate CMS configuration error about five days earlier exposed about 3,000 internal files. The outlet said details related to the unreleased Mythos model also leaked at the time.

The string of security incidents is fueling questions over the operating and security systems of the company, which is valued at $350 billion.

The outlet also reported speculation that Anthropic is considering an IPO in the fourth quarter of 2026, saying the repeated incidents could stand out as a risk factor in the listing process. It pointed to prevention measures for the open-source ecosystem (npm) distribution pipeline, steps to restore trust among enterprise customers, and the effectiveness of legal and technical responses to the spread of derivative projects based on the leaked code.