Anthropic leaks Claude Code source code: 500,000 lines exposed

Anthropic experienced a significant data leak on April 1, 2026 when parts of the source code for its AI coding tool Claude Code became publicly accessible due to human error.

The leak, caused by a configuration mistake, exposed nearly 2,000 files and 500,000 lines of code. It was the second leak in a short period: previously, a draft blog post was accidentally published revealing the existence of two unreleased models internally known as "Mythos" and "Capybara."

Anthropic confirmed the incident and characterized it as a configuration error with no malicious intent. The company is investigating the scope and has taken measures to prevent similar occurrences.

The leak came the day after Forbes reported on the Mythos model, which according to internal documents surpasses Claude Opus 4.6 on coding, reasoning, and cybersecurity. Anthropic is currently offering limited early access to Mythos due to the model's advanced cyber capabilities.

For enterprises relying on Claude as a secure AI partner, the leak raises questions about Anthropic's internal control procedures.