ANTHROPIC
Uh oh! Claude source code is now public
The blueprint for Anthropic's most-used AI tool, Claude, accidentally went public.
Over 512,000 lines of code and nearly 2,000 source files were leaked during the Claude Code update last week.
And because the internet is quick to spot glitches and slow to let people off the hook, it was immediately posted to X. From there, GitHub users hit Ctrl+C to save the code to their own repositories … over 50,000 times.
Yes, there are now 50K copies of Claude’s code.
But what was found in that leak, you may ask?
Developers spotted a Tamagotchi-style companion feature. Ie: a little pet that sits beside your input box and reacts to your coding in real time. They also found traces of something called ‘KAIROS’, a background agent that’s always on and can work on your behalf, even when you're not actively using the tool.
Here’s what you should know:
The source code leak revealed Claude’s structural DNA, along with updates that hadn't been announced yet.
According to Anthropic, this was not a security leak. Just a good old-fashioned human error.
This was the second leak within a week. The first exposed thousands of internal files.
Reputation vs. Reality
The Thursday before the source code slip-up, Anthropic accidentally released around 3,000 internal files. Among those documents was a blog post that previewed an unreleased, next-generation Claude model.
Anthropic’s response to the most recent faux pas was this measured press statement: “Earlier today, a Claude Code release included some internal source code. No sensitive customer data or credentials were involved or exposed.”
Okay, good to know user information is safe. But the tension here is that Anthropic’s brand hinges on being the responsible AI company. You know, the one that’s careful with data and prioritizes security above all else.
Oof, it’s a bad time to be Anthropic’s PR team.
Behind every AI company is a very human team - TL