Anthropic leaks source code of revenue-generating Claude Code AI

Anthropic has exposed 1900 files and 512000 lines of internal TypeScript code for its Claude Code AI tool through a public npm package update.The leak included the full architecture of the tool, which runs at a $2.5 billion annual revenue rate, plus unreleased features like an always-on Kairos agent and user frustration tracking. Copies appeared on GitHub before takedowns.This marks the second such leak in days, stemming from manual steps in their deployment process.

AI coding agents like Claude Code operated as black boxes, with their designs protected and hard for rivals to copy.The public blueprint now lets competitors dissect and improve on it fast, while highlighting fixable flaws like error-prone deploys. Builders gain a rare view of what powers billion-dollar tools, down to the frustration metrics.