Claude Code: Why Anthropic Reports Massive GitHub Repo's
Anthropic Is Reporting Lots of Git Repositories Claiming Claude Source Code Leak – But Here's the Real Messy Story Behind It All
This whole situation with Anthropic and Claude has gotten completely wild in the last couple weeks. You’ve probably seen the headlines or the YouTube videos popping up everywhere saying “Claude source code just leaked” or “check out these GitHub repos with the full Claude Code inside.”, at first it really did feel like one of those classic tech rumors that spread like crazy on Twitter and Reddit, repos were popping up overnight claiming they had the entire thing, and everyone from developers to AI hobbyists was cloning them like mad to see what was inside. But then things got even messier because Anthropic themselves started reporting and going after a ton of those Git repositories, which only made the whole thing blow up even more. So today I’m gonna walk you through exactly what’s been happening, how it started as what looked like rumors, and why it turned into this massive thing where thousands of repos got caught in the crossfire. No hype, no clickbait, just the straight story from what actually went down.
It all kicked off around the end of March 2026 when some security researcher named Chaofan Shou. He said something like “Claude code source code has been leaked via a map file in their npm registry.” And right away, people started digging. See, Anthropic had just pushed out version 2.1.88 of their Claude Code CLI tool to npm, and buried in that package was this huge 59.8 megabyte source map file that nobody was supposed to include. Source maps are normally just for debugging – they help map the minified production code back to the original readable TypeScript. But in this case, because of a simple packaging mistake, that map file pointed straight to a full ZIP archive sitting on Anthropic’s own cloud storage with basically the entire client-side codebase inside. We’re talking around 512,000 lines of clean, unobfuscated TypeScript spread across nearly 1,900 to 2,300 files. It wasn’t the model weights or any training data, thank goodness, but it was the full harness – the agent runtime, the multi-agent orchestration, the permission systems, the memory compaction stuff, all of it. And once that link went public, it spread insanely fast.
Now, here’s where the rumor part comes in that a lot of those early videos were focusing on. Right after the discovery, a bunch of GitHub repositories started appearing out of nowhere. Some had over 2,000 stars in just a few days, others claimed to have the full leaked code, and a lot of them were uploaded literally within the last three or four days before people started talking about it. You had one official-looking Anthropic repo that was already public for Claude Code stuff, but then tons of third-party ones popped up – some in Chinese accounts, some mirroring the code, some even porting it to Python or Rust using AI tools so they could dodge any takedowns. A few of them included source map files or partial TypeScript snippets, and creators were encouraging everyone to clone and run them locally to “verify” what was inside. But a lot of the early analysis videos were super skeptical. They were like, “hold up, is this actually a real leak or just people capitalizing on the hype?” Because at first glance, it was hard to tell if these repos had the genuine full source or just fragments, templates, or even fake stuff thrown together to ride the wave. The videos pointed out how easy it is for rumors to explode in the AI community, especially around something as big as Claude Code, which is this premium $100-a-month coding agent that lots of devs were already using heavily.