Anthropic Scrambles to Contain Claude Code Leak, Revealing Unannounced Features

Daily Technology

02/04/2026

Artificial intelligence firm Anthropic is reportedly in a state of disarray following a significant leak of its Claude Code source code. The exposed data not only details current functionalities but also reveals upcoming features, including a "Tamagotchi"-style virtual pet and a persistent, always-on AI agent named Kairos. The company is aggressively attempting to remove copies of the code from platforms like GitHub and address security vulnerabilities, indicating a strong desire to control the narrative and protect its competitive edge.

Key Takeaways

Anthropic's Claude Code source code was leaked, exposing unannounced features.
The leak includes details on a "Tamagotchi"-like virtual pet and a 24/7 AI agent called Kairos.
Anthropic is reportedly issuing thousands of copyright takedown requests to manage the leak.
The exposed code could provide competitors with insights into Claude Code's operations.

The Unveiling of Claude Code's Secrets

In a major security misstep, the source code for Anthropic's AI coding assistant, Claude Code, was discovered in a publicly accessible database. This leak has provided an unexpected glimpse into the company's development pipeline, revealing features that have yet to be officially announced. Among the most surprising revelations are plans for a virtual pet, reminiscent of the popular "Tamagotchi" game, and a persistent AI agent codenamed Kairos.

Kairos: The Always-On AI Agent

The leaked code details Kairos, an AI agent designed to operate continuously in the background, proactively assisting users. This persistent nature aims to make Claude more akin to advanced, open-source AI agents. A notable feature of Kairos is "autoDream," which reportedly allows the AI to consolidate and update its internal memories overnight, enhancing its long-term operational capabilities.

Anthropic's Damage Control Efforts

Sources indicate that Anthropic is in a state of panic, working rapidly to mitigate the fallout from the leak. The company has reportedly initiated over 8,000 copyright takedown requests on GitHub to remove copies of the Claude Code source code. While Anthropic attributes the leak to human error rather than a breach, the exposed code presents potential security risks, allowing malicious actors to probe for exploits. Furthermore, the leak offers competitors a clearer understanding of Claude Code's architecture, potentially enabling them to replicate its functionalities more easily.

Strategic Leaks or Security Failures?

This incident follows a pattern of recent leaks from Anthropic, including the discovery of plans for a new model called Mythos in a public database last month. Some speculation suggests these leaks could be strategic, aimed at generating investor interest ahead of a potential initial public offering. However, the reported urgency and panic within Anthropic suggest that the company would prefer this information to remain private, at least for the time being.

2026-03-23

Beyond Default Keyboards: A Technical Comparison of Mobile Input Methods

An objective comparison of top third-party keyboards like Gboard, SwiftKey, and others. Explore their technical features, AI capabilities, and privacy aspects.