Anthropic Leaks Shed Light on Hidden Features of Claude AI Model

Anthropic Leak Reveals Ambitious AI Features

A recent incident involving Anthropic has unexpectedly shed light on a set of unreleased features for its AI coding assistant, Claude. The leak, described as a "release packaging issue caused by human error," exposed internal source code that included feature flags for tools still under development. While Anthropic insists no sensitive user data was compromised, the leak has sparked intense interest across the developer community and raised fresh questions about operational security.

The leak revealed four secret features: BUDDY, a gamified AI companion; KAIROS, an always-on autonomous Claude; ULTRAPLAN, deep work in the cloud; and COORDINATOR MODE, multi-agent collaboration. These features suggest a future direction for AI systems that are more autonomous, persistent, and collaborative.

The 4 Secret Features Revealed

Read also: Google Pixel Watch Leak: Diver's Discovery Sparks Speculation

The BUDDY feature is expected to be teased around April 1, with a broader launch targeted for May 2026. KAIROS points to a major leap in autonomy, with a "dreaming cycle" that consolidates memory and improves performance overnight. ULTRAPLAN targets high-level problem-solving, enabling users to review, approve, or reject output, and "teleport" plans directly into a local terminal. COORDINATOR MODE is the most technically ambitious feature, allowing one instance of Claude to spawn multiple sub-agents working in parallel.

The Anthropic code was accidentally included in a routine update and published to a developer-facing registry. It was quickly identified and traced to a cloud-hosted archive containing a vast codebase, which was subsequently mirrored and analyzed by developers on platforms like GitHub. The rapid spread of the material allowed outsiders to piece together not just isolated features but a broader roadmap of Anthropic's ambitions.

Anthropic has stated that no sensitive user data or credentials were exposed in the leak. "Earlier today, a Claude Code release included some internal source code. No sensitive customer data or credentials were involved or exposed," according to a company spokesperson cited by Axio.

Read also: Chinese Startup Claims AI Coding Capabilities Exceed Those of GPT-5.5: A Detailed Examination

The incident raises questions about operational security at Anthropic, particularly given its positioning as a safety-focused AI lab. While the company has moved quickly to contain the fallout, the exposure has already given competitors and developers a detailed look at how advanced AI systems are being built and where the technology is headed next.