Massive Leak Exposes Claude AI Source Code to Public View

Post Views: 8

Mistake Exposes Secrets of Lucrative AI Coding Tool

On 31 March 2026, a critical blunder occurred when the AI powerhouse Anthropic inadvertently leaked the confidential blueprints for its highly profitable coding tool, Claude Code. The accidental disclosure happened through a routine update on the public npm registry, which resulted in the exposure of 512,000 lines of code.

What Happened?

The incident began when a 59.8 MB source map file was included in the update, making it possible for developers to translate complex computer code into plain English. This file was initially spotted by Chaofan Shou, an intern at Solayer Labs, at 4:23 am ET.

According to Anthropic, the leaked code generates $2.5 billion in yearly revenue, accounting for a substantial portion of the company’s estimated $19 billion total revenue.

The company has confirmed that no sensitive customer data was exposed in the breach. However, the leak poses a strategic threat, as it provides competitors with a comprehensive understanding of Claude Code’s functionality and underlying technology.

Analysis of Leaked Files

Further analysis of the leaked files reveals that Anthropic has developed a three-layer memory system to address context entropy, a common issue affecting AI systems. This system ensures that the AI does not become confused during extended tasks.

KAIROS: an always-on mode
Capybara (Claude 4.6): a future model
Fennec: another upcoming model
Buddy: a terminal pet

The leak coincided with a separate attack on a tool called Axios, which may have compromised some users’ computers if they updated via npm between 00:21 and 03:20 UTC on 31 March. To mitigate this risk, Anthropic is advising users to switch to the Native Installer directly from the website.