Back to News
NewsInflationInterest RatesFederal Reserve

Anthropic Discontinues Hidden Tracker in Claude Code Following Privacy Issues

Anthropic has removed hidden tracking from Claude Code after researchers found it identified users, particularly linked to China, raising privacy concerns.

Mercation Admin
July 7, 2026 2 min read
Anthropic Discontinues Hidden Tracker in Claude Code Following Privacy Issues

Anthropic Removes Covert Tracking from Claude Code

Anthropic has discontinued a hidden tracking system in its Claude Code AI assistant following the discovery that it was used to identify specific users, particularly those linked to Chinese AI labs. This decision comes after a security researcher revealed the existence of these undisclosed markers, which could pinpoint user locations and proxy usage.

The tracking feature, exposed by developer “Thereallo” in June, was intended to help prevent account misuse and identify potential AI model distillation attempts. Thereallo noted that the tracking included signals within Claude Code's prompts that indicated users suspected of circumventing restrictions.

While Thereallo recognized the rationale behind identifying unauthorized resellers and gateways, he criticized the lack of transparency in how these tracking signals were implemented, stating that the use of Unicode markers and encoded domain lists was inappropriate for a developer tool reliant on trust. "This is not a malicious feature, but it is an odd choice for a developer tool that relies on trust," he commented.

In response to the criticism, Anthropic engineer Thariq Shihipar explained on X that the tracking system was introduced in March as a temporary measure to combat account abuse and safeguard against distillation attacks. He confirmed that the company had strengthened its security measures since then and intended to remove the feature, which will be fully rolled back in an upcoming release.

This incident highlights Anthropic's ongoing concerns regarding AI model distillation, a practice where the outputs of one AI system are used to train another. The company has recently intensified its warnings about this practice, particularly in light of geopolitical tensions. Earlier this month, Alibaba labeled Claude Code as “high-risk” and restricted its employees from using the tool due to security concerns.

In February, Anthropic accused several Chinese AI companies of using fraudulent accounts to extract millions of responses from Claude for training their own models, a move that raised questions about industry practices. In April, Elon Musk testified that xAI had partially used OpenAI models for its own training, suggesting that distillation is a common industry practice. Anthropic CEO Dario Amodei has since urged Congress to enhance protections against foreign extraction of AI.