Anthropic's Claude Code leak due to build error, no user data at risk

A serious incident occurred at Anthropic when internal code for the Claude Code tool was accidentally made public. The leak happened due to an error during the build of version 2.1.88, which resulted in an unencrypted file, roughly 60 MB in size, containing instructions for AI agents being released onto the internet.

After discovering the problem, the company quickly filed DMCA takedown notices on GitHub, successfully removing over 8,100 repositories that contained the leaked materials. Anthropic emphasized this was not a hack, but a human error in product packaging, and that no user data was compromised.

Meanwhile, the developer community attempted to circumvent the restrictions. Enthusiasts used neural networks to rewrite the leaked code, altering its structure to avoid automated blocks. This allowed some of the material to continue circulating even after the mass deletions.

Interest in the leak intensified after the file was found to contain references to internal anti-copying mechanisms and unannounced features. These included an autonomous mode called KAIROS and a system referred to as an "AI companion," reminiscent of a digital pet. However, some users believe these findings are experimental or even a joke.