Anthropic accidentally leaked Claude code source code via npm slip-up
Anthropic accidentally leaked the source code for its AI agent, Claude Code, thanks to a packaging slip-up in an npm package.
The leak, spotted by Chaofan Shou on March 31, 2026, includes more than 500,000 lines of code.
Even though Anthropic quickly sent out takedown notices and called it human error, the code is still spreading across the internet.
Leak shows unreleased Kairos and Buddy
The files reveal some unreleased features like Kairos (an always-on memory agent), Buddy (a companion system with different species), and a massive React component.
These details are giving developers everywhere, especially in China, an inside look at Anthropic's unique coding style and tech secrets.
The whole thing has sparked big conversations about how companies protect their AI innovations as competition heats up.