A standard software update for Claude Code unexpectedly disclosed details about upcoming enhancements that Anthropic is developing. According to coverage from Ars Technica, The Verge, and various outlets, the release of version 2.1.88 on Tuesday included a file revealing the application's underlying source code. Prior to Anthropic's efforts to contain the exposure, the code was shared on a public GitHub site and downloaded over 50,000 times. In the end, more than 512,000 lines of code across 2,000 TypeScript files became accessible to the public and rival firms.
Following the incident, individuals reported discovering indications of new functionalities under development at Anthropic. On the platform X, Alex Finn, who founded the AI company Creator Buddy, indicated he spotted a setting for 'Proactive' mode, enabling Claude Code to operate without explicit user instructions. Finn also mentioned traces of a cryptocurrency payment mechanism that might enable AI agents to handle transactions independently. Additionally, a Reddit discussion highlighted by The Verge suggested Anthropic explored a virtual pet similar to Tamagotchi that responds to programming activities, possibly as an April Fools' concept.
An Anthropic representative stated to BleepingComputer: 'A Claude Code release contained certain internal source code. No sensitive customer information or access credentials were compromised or revealed.' The official added: 'This resulted from a packaging error during release due to human oversight, rather than a security violation. We are implementing steps to avoid future occurrences.'
Leaks like this remind us that development intentions frequently evolve. The presence of code for a specific function does not guarantee its public release by the company.