Anthropic removes 8,000 copies of 'AI disaster' it blamed humans for

Anthropic is scrambling to contain the leaked internal instructions that power Claude Code, its popular AI coding assistant. According to a report by The Wall Street Journal, the company has been successful in forced removal of 8,000 copies of the code by using a copyright takedown request. The leak was the result of a mistake that may cost company dearly. On Tuesday, while rolling out an update to Claude Code, Anthropic mistakenly published a file to GitHub that linked directly back to the tool's underlying source code - instructions that are hidden to prevent outside access. An X (formerly Twitter) user spotted the exposed file almost immediately and spread the word across the platform. Within hours, copies were multiplying across GitHub and Anthropic was quick to clarify the limits of the damage. The company later confirmed that the leak did not expose any customer data or personal information. Crucially, it also did not reveal the AI model weights and it blamed 'human' for the mistake.However, the leaked data was a fascinating for developers who were able to learn how Anthropic engineers work to let its AI systems to behave. Among the discoveries generating the most buzz online:Dreaming: A feature that periodically instructs Claude Code to pause, review its previous tasks, and consolidate its memories.Going Undercover: Instructions that appear to tell Claude Code to, in certain situations, conceal the fact that it is an AI when publishing code to platforms like GitHub.Future product hints: Developers found tags in the code that appear to point toward unannounced product releases, sparking immediate speculation about what Anthropic has in the pipeline.Buddy: Buried in the code was a Tamagotchi-style virtual pet named Buddy that users could apparently interact with.The leak exposed roughly 2,200 files and 30MB of TypeScript. According to engineers who dug through the code, this is at least the third time Anthropic has made this exact mistake.This comes at a time when reports say that Anthropic is eyeing a $380 billion IPO later this year. Bloomberg reported this week that Anthropic is in early discussions with Goldman Sachs, JPMorgan, and Morgan Stanley about a potential October IPO. The company has already rattled markets this year-its Cowork and Claude Code Security updates wiped billions from software and cybersecurity stocks in a matter of weeks.