Anthropic, a startup, has confirmed a leak of part of the source code for its AI programming tool, Claude Code, as reported by CNBC.
“No confidential client data has been exposed. The error occurred during the release packaging. It was caused by human error, not a hack. We are implementing measures to prevent this from happening again,” a company representative stated.
The first reports of the leaked code surfaced on March 31. A related post on X garnered over 30 million views.
Claude code source code has been leaked via a map file in their npm registry!
Code: https://t.co/jBiMoOzt8G pic.twitter.com/rYo5hbvEj8
— Chaofan Shou (@Fried_rice) March 31, 2026
The incident could negatively impact Anthropic’s standing, as it opens access to the algorithms of the popular tool to external developers and competitors.
The published data fully reveals the architecture of Claude Code. Specifically, it became known how the service:
- manages requests to models;
- coordinates the work of multiple assistants;
- controls access and permissions;
- handles authorization.
Additionally, the code revealed 44 hidden features that had not yet been officially presented. Among the most intriguing discoveries are:
- Kairos — a constantly active background system. It retains “memory” and analyzes accumulated data at night for further training;
- Buddy — an AI pet in the style of “Tamagotchi.” According to the code, its launch was planned for April 1-7.
The data set also includes Undercover Mode — a special mode that prevents the neural network from accidentally publishing internal project names.
A Series of Leaks and Market Competition
This is the second major error by the company recently. Just days prior, the description of a future AI model and other documents were found in the public domain, as reported.
At that time, a representative of Anthropic stated that the new model represents a “qualitative leap” in performance and is “the most powerful solution to date.” It is currently being tested among a limited group of users.
The company explained that information about the upcoming LLM appeared online “due to human error,” and the leaked materials were “early drafts of content.”
The release of Claude Code to the general public took place in May 2025. The tool assists developers in creating functions, fixing bugs, and automating tasks.
Over the past year, the service has become so in demand that growing competition from Anthropic forced OpenAI to shut down the Sora project to concentrate resources on developing similar solutions.
In March, Anthropic transformed Claude into an AI agent — the bot gained the ability to use a computer to perform tasks.