Claude Code leak exposes a Tamagotchi-style ‘pet’ and an always-on agent
Why it matters: The leak of over 512,000 lines of Claude Code's source code could provide bad actors with avenues to bypass Anthropic's AI guardrails.
- Anthropic's Claude Code 2.1.88 update accidentally included a source map file containing its TypeScript codebase, as reported by The Verge AI and Ars Technica.
- Users on X and Reddit quickly identified and shared the leaked code, which reportedly contains over 512,000 lines, offering a deep look into the AI tool's internal workings.
- Leaked code appears to show unreleased features, including a Tamagotchi-like 'pet' that interacts with coding input and a 'KAIROS' feature for an always-on background agent, according to user discoveries.
- Anthropic spokesperson Christopher Nulty confirmed to The Verge that the leak was a 'release packaging issue caused by human error,' not a security breach, with no sensitive customer data exposed.
- Arun Chandrasekaran, an AI analyst at Gartner, told The Verge that while the leak poses risks like enabling bad actors to bypass guardrails, its long-term impact might be limited to prompting Anthropic to improve operational maturity.
- The leaked code was copied to a GitHub repository that has since amassed over 50,000 forks, despite Anthropic fixing the original issue.
A recent update to Anthropic's Claude Code exposed over 512,000 lines of its TypeScript codebase, revealing unreleased features like a Tamagotchi-style AI pet and an always-on background agent named KAIROS, according to reports from Ars Technica and VentureBeat. While Anthropic quickly fixed the packaging error, the leaked code has been widely copied on GitHub, prompting concerns about potential misuse despite the company's assurance that no sensitive data was compromised.
