The leaked code, over 512,000 lines, reveals unreleased features, instructions for Claude, and more. Following Anthropic’s release of the Claude Code 2.1.88 update, users found a source map file with its TypeScript codebase, shared online by some, including one notable instance on X. The leaked data, reported by Ars Technica and VentureBeat, shows the inner workings of the AI coding tool.
Users exploring the code found potential new features, Anthropic’s instructions for the AI bot, and its “memory” architecture. Highlights include a Tamagotchi-like feature reacting to input, a “KAIROS” background agent, and a coder’s comment admitting increased complexity due to memoization.
Though Anthropic quickly corrected the issue, users replicated the code on a Github repository, gathering over 50,000 forks. Launched in February 2025, Claude Code gained traction by adding agentic capabilities.
“Earlier today, a Claude Code release included internal source code. No sensitive customer data or credentials were involved or exposed,” Anthropic’s spokesperson, Christopher Nulty, clarified in a statement, attributing the lapse to human error, not a security breach, with measures underway to prevent recurrences.
AI analyst Arun Chandrasekaran from Gartner told The Verge that while the leak might provide bad actors a path to bypass guardrails, its long-term impact could prompt Anthropic to enhance operational maturity processes and tools.