Anthropic accidentally leaks Claude Code

Anthropic inadvertently published internal Claude Code source in a public npm release, exposing about 500,000 lines of code that reveal internal memory architecture, an autonomous "KAIROS" feature and the company's AI roadmap (Capybara, Fennec, Numbat). The company says no customer data or credentials were leaked and attributes the exposure to a release packaging error, but the disclosure risks IP loss and could help adversaries understand and attempt to bypass safety controls.