Claude Code: Massive Leak Reveals a Secret Agent and a Tamagotchi

A massive leak of Claude Code, an AI-powered coding tool developed by Anthropic, has revealed intriguing and unpublished features. Over 512,000 lines of code were accidentally made available to users following the 2.1.88 update. This leak was initially reported on the X platform, where a user shared a file containing the TypeScript code of the tool. Media outlets Ars Technica and VentureBeat also reported on the incident.

Users who explored the code discovered novel features, including a Tamagotchi-style virtual pet that interacts with the coding process. Another feature, named “KAIROS,” could allow an agent to operate continuously in the background. These findings were shared on Reddit, where users also found comments from Anthropic developers. One notable comment mentions that memorization significantly increases complexity, and it is uncertain whether it truly enhances performance.

In addition to these features, the leaked code contains instructions from Anthropic for the AI bot and information about its memory architecture. Despite Anthropic's swift correction of the error, the code has been widely disseminated, with over 50,000 copies made on GitHub. Claude Code, launched in February 2025, has gained popularity due to its task automation capabilities.

Christopher Nulty, a spokesperson for Anthropic, clarified that this leak did not involve any sensitive customer data and resulted from human error during the packaging of the release. Anthropic has since taken steps to prevent such errors in the future.

Arun Chandrasekaran, an analyst at Gartner, emphasized that while the leak may present opportunities for malicious actors, it could primarily prompt Anthropic to enhance its security processes and operational maturity. The leak of Claude Code poses risks such as providing malicious actors with opportunities to bypass safeguards.