With the source of Anthropic's key development tool "Claude Code" leaking online, concerns are rising that exposure of trade secrets could also weigh on its initial public offering (IPO) plans.
According to the security industry on the 31st (local time), the Claude Code source was made public through NPM, a developer package repository, and then quickly spread to multiple platforms including GitHub.
This leak is presumed to be a mistake in the internal distribution process, not a hacking incident. The cause has been identified as "map files," which should not have been included during code distribution, being released together, making it possible to restore encrypted code to its original form.
The leaked volume reportedly exceeds 510,000 lines, with about 1,900 files. However, core AI models such as Claude Opus, Sonnet, and Haiku were not included in the leak.
Experts note that the risk is high because this incident could reveal not only source code but also development methods and internal know-how to outsiders.
As Anthropic has grown by focusing on the enterprise AI market, development tools like Claude Code are considered central to its competitiveness. For this reason, there is speculation that the incident could affect the IPO being prepared for the fourth quarter of this year.
Earlier, on the 25th, Anthropic also experienced a security issue in which information related to its next-generation model "Claude Mythos" was exposed due to an internal system configuration error, one of a series of recent security problems.