Anthropic Leaked Its Own Source Code, Then Accidentally Nuked 8,100 GitHub Repos Trying to Clean It Up

On April 1st, Anthropic shipped a routine update to Claude Code. Buried in that release was something that was never supposed to see daylight: the full 512,000-line source code for their most popular product.

A developer spotted it. Within hours, a repository called “claw-code” had 50,000 stars. That makes it the fastest-growing repo in GitHub history. Developers weren’t just starring it. They were forking it, dissecting it, and porting it to work with GPT, Llama, and every other model they could get their hands on.

Anthropic’s response? Mass DMCA takedowns. And this is where it gets ugly.

In the scramble to pull leaked code off GitHub, Anthropic’s takedown requests were so broad that approximately 8,100 repositories got caught in the crossfire. Projects that had nothing to do with Claude Code. Open source tools people depended on. Gone overnight because someone’s regex was too greedy on the DMCA filing.

By April 2nd, Anthropic walked it back. Called it an accident. Repos started coming back online.

But the damage was already done in more ways than one. The Streisand effect kicked in hard. Every takedown notice became a news story. Developers who might have glanced at the code and moved on were now actively rewriting it in Rust, Go, and Python specifically to create versions that couldn’t be DMCA’d because they weren’t direct copies.

There’s a real lesson here for every company building with AI. Your source code is one bad deploy away from being public. And your incident response can easily do more damage than the incident itself.

Anthropic built one of the best AI coding tools on the market. Then they showed the world exactly how it works, panicked, and broke thousands of projects that had nothing to do with them.

The code is out there now. It’s not going back in the bottle. The only question is whether Anthropic learns from the response or doubles down.