Anthropic Removed Hidden Anti-Distillation Code from Claude Code After Its Discovery

On July 1, 2026, Anthropic updated Claude Code — its AI-powered development tool — by removing hidden code that the company had used for several months to identify developers attempting to steal data from its models through distillation, including targeting Chinese AI companies.

What the Hidden Code Did

The code fragment was added to Claude Code several months ago without any public notice: tool users neither knew of its existence nor its function. The mechanism was aimed at detecting unauthorized model distillation — a practice in which competitors send massive requests to Claude, collect responses, and use them as a training dataset for their own language models.

Distillation of knowledge itself is a legal machine learning technique: a more powerful "teacher" model transfers behavioral patterns to a more compact "student" model. This method allowed a number of Chinese developers to create competitive models with significantly lower costs. However, systematic extraction of knowledge from a closed commercial API without permission violates the terms of service and is de facto theft of intellectual property.

Key facts:

• Hidden code added several months ago — without user notification
• Function — to identify developers using Claude for illegal distillation
• Presumed targets — Chinese AI companies
• Removed on July 1, 2026 after discovery and public disclosure
• Anthropic confirmed the fact of code existence and its removal

Why Anthropic Removed the Code After Public Disclosure

When information about the hidden mechanism became public, Anthropic found itself in a contradictory position: a company consistently promoting the principles of safe and transparent AI was using hidden tracking tools inside its own developer product. In response, Anthropic removed the code in the July 1, 2026 update and publicly acknowledged its existence without revealing details about the scale of use and data collected.

It is precisely this aspect that raises the most serious questions: engineers using Claude Code as a work tool did not suspect that they were simultaneously participating in a corporate counter-intelligence operation. The line between business protection and abuse of user trust turned out to be thin.

What This Means

The incident exposes an escalating confrontation in the global AI race: leading Western labs — Anthropic, OpenAI, Google DeepMind — are experiencing competitive pressure from Chinese companies producing powerful models at significantly lower costs. Suspicions of systematic distillation of Western models have long been heard in the industry. But the Claude Code story shows: defensive measures, if hidden even from the clients themselves, can inflict no less reputational damage than the threat they attempt to protect against.