AI: Anthropic Removes Claude Code's Hidden Tracker After Controversy

Anthropic removed a hidden tracking system of Claude Code after the discovery, in June, of markers identifying certain Chinese users. The company cites the fight against account abuse and model distillation. Is AI tools transparency still compatible with this form of surveillance?

What does the discovery of the hidden tracker reveal?

The researcher Thereallo uncovered, in June, a hidden code in the system prompts of Claude Code. This mechanism flagged users suspected of circumventing Anthropic’s restrictions or extracting the model’s capabilities.

The system relied on Unicode markers and encoded domain lists. A hostname containing “deepseek” or “zhipu” was, for example, a signal exploited by the tool.

Thereallo did not consider the feature malicious in itself. However, he criticized the complete lack of documentation or release notes on this point, for a development tool that relies on users’ trust.

Why was Anthropic monitoring certain users?

The engineer Thariq Shihipar confirmed, on X, that the experiment started in March. The goal was to prevent account abuse by unauthorized resellers and to limit distillation, the practice of training a competing model from Claude’s responses.

Furthermore, in February, Anthropic accused the Chinese developers DeepSeek, Moonshot AI and MiniMax of siphoning millions of responses via fraudulent accounts. Dario Amodei later mentioned in June nearly 25,000 accounts linked to Alibaba and 28.8 million suspicious transactions.

This controversy takes place in an already tense climate: Alibaba banned the use of Claude Code by its teams, considering the tool high-risk after this discovery. Another episode in the growing technological rivalry between the United States and China.

Anthropic therefore did not act in a void. However, the chosen method, undisclosed surveillance, weakened developers’ trust in the tool.

The patch should restore a standard operation of Claude Code, without a hidden marker. The issue of transparency of anti-abuse detection systems remains fully open for the entire generative AI sector.