An Anthropic engineer confirmed the removal of a secret tracker embedded in Claude Code, a developer tool powered by artificial intelligence. The hidden code, discovered by a security researcher, monitored Chinese users without their explicit consent. The revelation has sparked a heated debate about transparency in AI companies and the boundaries between legitimate protection and covert surveillance.
Prompt steganography used to conceal tracking mechanisms
Researcher known as "Thereallo" identified the presence of prompt steganography, a technique that hides data within seemingly normal messages. The invisible code collected information such as timezone, proxy, and potential connections to Chinese AI labs previously accused by Anthropic of conducting distillation attacks. The discovery occurred while the researcher was analyzing privacy issues in Claude Code, shocked at how easily the tracking was concealed.
Sponsored Protocol
Anthropic defends the experiment but the community reacts with outrage
Thariq Shihipar, an engineer at Anthropic, stated on X that the tracker was an "experiment" launched in March to prevent abuse by unauthorized resellers and protect against distillation. According to the Washington Post, unauthorized sellers offered access to free models for $1 per month and pro subscriptions worth $100 for as little as $12. Despite these justifications, the community condemned the lack of transparency, calling it a "serious breach of user trust." This incident echoes other surveillance cases in the tech industry, such as the complex legal battles between Apple and Epic Games where platform transparency was at the forefront.
Implications for user trust and regulation
The episode raises crucial questions about AI regulation and corporate responsibility in openly disclosing monitoring practices. While Anthropic argues the measure was necessary to combat fraud and intellectual property theft, the use of covert techniques undermines developer trust in Claude Code for sensitive projects. The removal of the tracker is a first step, but the community demands stronger guarantees. Events like this could influence future European and international regulations on surveillance in AI systems, similar to privacy violation sanctions. For more on platform transparency issues, see the article on the Apple-Epic Games case. Additionally, recent revelations about Samsung chip performance show how user trust can be strained by unclear corporate decisions. Lastly, steganography is an ancient technique, but its use in AI remains poorly regulated, as explained on Wikipedia.
Sponsored Protocol
