Anthropic accidentally took down 8,100 GitHub repos to scrub leaked Claude Code source

If you want to know the secret sauce behind the best AI coding assistant in the world, you usually have to guess. Anthropic just gave it away by accident.

On Tuesday, a software engineer noticed that Anthropic had accidentally bundled the full source code and underlying instructions of its flagship Claude Code CLI tool into a recent public release. Naturally, the AI community jumped on the opportunity, reverse-engineering how Anthropic directs its models and re-hosting the files all over GitHub.

Then came the panic button, followed by an even bigger mess.

When a cleanup takes down half the building

Anthropic scrambled to contain the leak by filing a copyright takedown request under U.S. law. But when GitHub executed the notice on Wednesday morning, it took down roughly 8,100 repositories in one sweep.

That list included thousands of legitimate developer forks connected to Anthropic’s publicly released Claude Code project. Frustrated developers took to social media to voice their anger after finding their own innocent projects blocked.

Boris Cherny, head of Claude Code at Anthropic, admitted the mass takedown was a mistake. The company quickly pulled back the notices, shrinking the scope to just one primary repository and 96 forks that actually held the leaked source code.

An Anthropic spokesperson explained to TechCrunch that the targeted repo was part of a larger fork network tied to its public repository, which caused the takedown to spread much wider than planned. A spokesperson also told The Wall Street Journal that no customer data was exposed, nor were the valuable weights of Anthropic’s underlying models leaked. The company categorized the incident as a packaging issue caused by human error rather than a security breach.

Why it matters

Claude Code has become a benchmark in the AI coding space, meaning developers will spend weeks dissecting the leaked scaffolding to see how frontier models are actually prompted and orchestrated under the hood. If you build AI agents, the industry just got a rare look at the blueprint.

However, the cleanup effort looks remarkably sloppy. Deploying a sweeping automated DMCA hammer that knocks out 8,100 repositories demonstrates how fragile developer trust can be when big platforms overreach.

With Anthropic reportedly eyeing an eventual IPO, we think this incident highlights a glaring operational blind spot. It is one thing to suffer an embarrassing packaging bug, but quite another to disrupt thousands of active developers in an attempt to bury it. Execution and compliance will matter just as much as raw intelligence once public markets enter the picture.