Anthropic, the company behind the Claude AI assistant, has apologized after being caught running an undisclosed experiment that tracked users without their knowledge.
According to Ars Technica, Anthropic was "outed" for a Claude tracker that secretly monitored Chinese users. The report says users were shocked to learn of the surveillance, particularly because it appeared to run counter to Anthropic's public anti-surveillance stance.
Ars Technica reports that the company was accused of spying on users, and that an Anthropic engineer said the "experiment" is now over. The story surfaced more broadly after being picked up through Google News coverage citing the Ars Technica reporting.
The core tension here is one of trust. Anthropic has positioned itself as a safety-focused AI lab that emphasizes responsible development, so the revelation of a covert tracking program — described by the company as an experiment — sits uneasily with the image it projects. When users interact with an AI assistant, they often share sensitive questions, personal details, and work information, and they generally assume that data isn't being quietly monitored beyond what they've agreed to.
Based on the available reporting, key details remain limited: the sources do not spell out exactly what data was collected, how many people were affected, how long the experiment ran, or precisely how it was discovered. What is clear from Ars Technica is that the tracking was secret, that it drew accusations of surveillance, that it specifically involved Chinese users, and that Anthropic has both ended the experiment and apologized.
Why it matters: as millions of people fold AI assistants into their daily lives, this episode is a reminder that the companies building those tools can quietly experiment on the people using them — and that a stated commitment to privacy is only as meaningful as a company's actual conduct behind the scenes.