Anthropic enables independent AI research via privacy-protected data pilot

Anthropic has launched a pilot program to facilitate independent research on how people interact with its Claude AI model. To address privacy concerns, the company uses a system called Anthropic Insights, which provides external researchers with aggregated data rather than raw conversation transcripts.

Three research institutions participated in the pilot, analyzing approximately 250,000 conversations from the April to May 2026 period. Stanford University’s SALT Lab focused on human-AI collaboration, the University of Oxford’s Human Information Processing Lab examined the relationship between user experience and model behavior, and METR investigated productivity related to Claude Code.

Findings from the studies include observations that 56% of assessable Claude tasks were considered high-stakes or consequential. The program aims to bridge the gap between proprietary corporate data and public datasets, which often lack the professional or academic use patterns found in commercial AI interactions. A third-party privacy audit was conducted to ensure the security of the data-sharing mechanisms.