AI company Anthropic recently accidentally leaked 512,000 lines of its code for the tool Claude Code. What’s crazy about this situation is that it wasn’t a hack or an attack. IT WAS HUMAN ERROR!
This happened because they released an update for Claude and accidentally left a debug file in it, specifically a source map. That file allows people to see the original code, and when people downloaded the update, they saw the code, and BOOM, LEAKED.
The craziest part is that no one even had to hack anything, it was literally publicly available to everyone.
Almost 2,000 files were leaked and over 521,000 lines of code. However, no user data or passwords were leaked. Only the tool that communicates with the AI, specifically the CLI, was leaked.
This is a big problem for Anthropic because developers can now see how Claude Code works internally and discover hidden features that haven’t been released yet, such as a “self-healing memory” system, an AI agent that can run in the background, etc.
The competition is basically getting a “cheat sheet” other AI teams can see how the system is built and potentially copy it. They are literally getting a free blueprint.
It’s also a risk to their reputation, since Anthropic presents itself as an “AI safety company,” and a mistake like this doesn’t look good at all.
This spread when someone posted the link on X (Twitter), and people moved the code to GitHub. Thousands of people started analyzing it. They removed the code, but it was already too late.
Even though this isn’t a catastrophe for users, since no user data was leaked, it is a major blow for the company.
Discussion(0)