Support for Independent Research on Claude Usage

Anthropic conducted a pilot program to enable external researchers to conduct independent research using real user data. The goal is to enhance objective understanding of AI usage patterns and expand data accessibility.

Three institutions, including the Stanford SALT Lab, Oxford Human Information Processing Lab, and METR, participated and analyzed approximately 250,000 conversation records. Researchers handled the data using a privacy-preserving tool called Anthropic Insights, while Anthropic minimized intervention to prevent bias in the research results.

Key research findings confirmed that people tend to delegate more significant tasks to AI than expected. This tendency was particularly pronounced when seeking professional advice related to law and finance. Additionally, in most conversations, users did not use AI outputs as-is but instead played a role in setting direction and making corrections.

The aggregated data collected through this pilot has been made public, and it is expected to be utilized by more researchers and policymakers in the future to understand the societal impact of AI.