Anthropic opens Claude usage data to outside researchers

Anthropic said three external groups — Stanford's Social and Language Technologies (SALT) Lab, the University of Oxford's Human Information Processing Lab, and the evaluations non-profit METR — each designed their own studies and analyzed roughly 250,000 Claude.ai or Claude Code conversations from April–May 2026 through Anthropic Insights, the privacy-preserving analysis tool formerly called Clio. Anthropic ran the data collection; the groups did their own analysis. The company says this is the first time external researchers have run public independent studies on an AI company's own usage data, and it is releasing the aggregate data from each project along with an expression-of-interest form for future participants.

Anthropic says its contractual review rights were limited to user privacy, information that could aid usage-policy violations, its own confidential information, and research accuracy, and that researchers are free to publish inconvenient results. It also ran an additional privacy audit of everything shared with the third parties.

The SALT Lab found that over half of Claude conversations involved people delegating consequential tasks — work that affects others or is hard to undo — contrary to prior research suggesting users keep such work for themselves; professional guidance on legal and financial questions was the most common case. In nearly three-quarters of conversations, the human set direction and adapted rather than copied Claude's output. Oxford reported correlations between Claude's behavior and users' emotional states.

Only SALT's full writeup is public. Oxford's is unfinished, and METR's estimates of coding-agent productivity gains across model generations are not detailed in the post.