Claude Code Source Code Leaked: 44 Unreleased Features Exposed

Claude Code Source Code Surfaces Online

Anthropic has confirmed an accidental leak of its Claude Code source code, according to CNBC. A company spokesperson stated that no confidential customer data was exposed and attributed the incident to a human error during release packaging rather than a security breach. The startup pledged to implement safeguards to prevent similar occurrences.

Reports about the leaked code first emerged on March 31. A post by researcher Chaofan Shou on X quickly went viral, accumulating over 30 million views.

"Claude code source code has been leaked via a map file in their npm registry!" — Chaofan Shou (@Fried_rice), original post

Why This Matters

The leak fully exposed the internal architecture of Claude Code — one of the most widely used AI coding assistants on the market. Third-party developers and competitors now have visibility into how the tool manages model requests, coordinates multiple assistants, handles access controls and permissions, and processes authorization. This could significantly undermine Anthropic's competitive edge in the rapidly expanding AI coding assistant space.

44 Hidden Features: From Self-Learning to an AI Pet

Analysis of the published codebase uncovered 44 features that Anthropic had not yet publicly announced. The most notable discoveries include:

• Kairos — a constantly active background system that retains a "memory" of interactions and analyzes accumulated data overnight for model fine-tuning.
• Buddy — a virtual AI companion resembling a Tamagotchi-style pet. Based on the code, its launch was scheduled for April 1-7.
• Undercover Mode — a specialized mode designed to prevent the AI from accidentally disclosing internal Anthropic project names.

Second Leak in Quick Succession

This marks Anthropic's second such blunder in recent memory. Previously, a description of an upcoming AI model along with related documentation surfaced online. At the time, the company characterized the new model as a "quantum leap" in performance and "the most powerful solution to date." That model was reportedly undergoing limited user testing. Anthropic attributed the earlier leak to the same root cause — human error — and described the exposed materials as "early content drafts."

Claude Code and the AI Market Rivalry

Claude Code launched to the general public in May 2025. The tool enables developers to build features, fix bugs, and automate routine tasks. Over the past year, the service gained such traction that mounting competitive pressure from Anthropic forced OpenAI to shut down its Sora project in order to redirect resources toward developing rival offerings.

In March, Anthropic upgraded Claude into a full-fledged AI agent, giving the bot the ability to independently operate a computer to complete user tasks. Recurring leaks pose reputational risks for the company at a time of heightened competition in the AI solutions market.