Artificial Intelligence

Anthropic Leaks 512K Lines of Source Code

Was Anthropic's code leak an embarrassing safety failure or a calculated move to crowdsource AI feedback?

APR 1·Updated APR 3

Anthropic Leaks 512K Lines of Source Code

Story

Above: Anthropic's logo in Creteil, France, on March 27, 2026. Image credit: Samuel Boivin/NurPhoto/Getty Images

The Facts

Anthropic has confirmed that parts of Claude Code's internal source were inadvertently leaked after a release packaging error. "This was a release packaging issue caused by human error, not a security breach," a company spokesperson said, adding that no customer data or credentials were exposed.
Verge
CNBC
The Economic Times
The leak occurred when Claude Code v2.1.88 was published to the npm registry with a source map file, cli.js.map, that contained the full TypeScript source code. The file reportedly mapped 1,900 files and more than 512,000 lines of code.
Cybernews
The Register
GitHub
A post on X announcing the leak drew tens of millions of views, and copies of the code were mirrored to GitHub repositories, one of which accumulated more than 50,000 forks.
Cybernews
Business Insider
Chaofan Shou on X
Analysis of the leaked code revealed several unreleased internal features, including further details on AI model Mythos, a background always-on assistant mode, and a Tamagotchi-style companion system called "Buddy" with 18 species of varying rarity gated behind a compile-time feature flag.
GitHub
Verge
The source code also contained an "Undercover Mode" designed to prevent Anthropic employees from revealing internal information when using Claude Code on public repositories, instructing the AI not to include internal codenames or disclose that it is an AI in commit messages or pull requests.
GitHub
The Register
The incident marked Anthropic's second data exposure in under a week, following a separate leak of model descriptions.
CNBC
Verge
The Economic Times

Sources Split

LEFT

CENTER

RIGHT

ESTAB
CRIT

CENTER

PRO
ESTAB

The Spin

Narrative A

This is yet another embarrassing mishap for Anthropic. A brand built upon superior AI safety has leaked over 500,000 lines of internal code through a basic mistake. Before lecturing the rest of the world on safety and security, Anthropic should first fix the mess at home.

Medium

Tuki on X

Narrative B

This is nothing more than an unfortunate, one-off human error with no sensitive customer data leaked in the process. If one thing is to be taken from the incident, it is that Anthropic continues to build game-changing AI capabilities for its customers.

Verge

Ole Lehmann on X

Cynical narrative

This incident may be more than an accidental error. As major AI companies require increasing levels of data, Anthropic will be receiving unrivalled feedback from the millions currently exploring their "leaked" code. Just in time for April Fools, things may not be as they seem.

Minus Wells on X

Damian Player on X