Anthropic's source code for its Claude Code AI assistant leaked online, exposing internal work on a feature called Proactive mode that could keep the tool active without explicit user prompts. The breach, caused by a packaging error in the 2.1.88 update, allowed the code to be copied over 50,000 times before Anthropic removed it from public repositories. While the leak revealed potential features like autonomous crypto payments and a Tamagotchi-style coding companion, the company emphasized no customer data was compromised.

Experts caution that leaked code doesn't guarantee future features, as plans can change. The incident highlights risks in AI tool development and the potential for leaked features to influence market expectations.