Anthropic opens Claude usage data to outside researchers

Anthropic said three outside research teams have finished the first independent studies of an AI company's own conversation data, working through a privacy-preserving system the company calls Anthropic Insights.

The pilot, announced on August 26, gave researchers at Stanford's Social and Language Technologies (SALT) Lab, the University of Oxford's Human Information Processing Lab, and the nonprofit AI evaluator METR access to roughly 250,000 Claude.ai or Claude Code conversations from April and May 2026. None of the outside teams saw raw chats. They wrote questions that Claude answered across the dataset, and got category-level aggregates back.

The Stanford team reported that "over half of Claude conversations involved people delegating consequential tasks to AI." Oxford looked at how the emotional texture of a session tracked with user behavior. METR studied productivity gains from coding agents across Claude generations.

Anthropic's framing is that this is "the first time external researchers have run public independent studies on an AI company's own usage data." The company said its contractual review of the resulting papers was limited to four things: "user privacy, information that could help people violate its usage policies, Anthropic's confidential information, and research accuracy," with the teams otherwise free to publish results "even if inconvenient for Anthropic."

The aggregate data is posted on Hugging Face, and Anthropic has opened an expression-of-interest form for further partners. Neither the post nor the search results name which outside proposals, if any, have been accepted for a next round, or how often the four review grounds were invoked during the pilot itself.