Anthropic's Claude Code source code leak
Anthropic confirmed on Tuesday that it had accidentally published part of the internal source code for Claude Code, its AI-powered software engineering assistant, after a human error caused an internal-use file to be bundled into a software update. That file pointed to an archive containing nearly 2,000 files and 500,000 lines of code.
The material spread quickly. A post on X sharing a link to the leaked repository accumulated more than 29 million views by early Wednesday, and a rewritten version of the code became GitHub's fastest-ever downloaded repository. Anthropic responded with copyright takedown requests in an attempt to limit further distribution.
Developers who reviewed the code reported finding designs for an always-on AI agent and a Tamagotchi-style interactive coding assistant, according to The Verge, suggesting Anthropic has been developing product directions it had not publicly disclosed.
The incident is particularly sensitive given Anthropic's positioning as one of the more safety-conscious frontier AI developers. Exposure of internal system architecture raises questions about competitive vulnerability as well as the security practices governing proprietary model infrastructure. The company has not disclosed whether any model weights or security-critical components were included in the leaked files.
Separately, Fortune reported that Anthropic is testing a new model internally referred to as Mythos, described as its most powerful AI system to date. That report surfaced alongside the leak coverage but appears to stem from independent sourcing.
