Societal Impacts Research

Source: Anthropic Research Date: 2025-12-02 URL: https://www.anthropic.com/research/team/societal-impacts

Summary

Team overview page for Anthropic’s Societal Impacts research group, covering empirical studies of AI behavior in real-world deployment. Representative work includes analysis of 700,000 interactions to build a taxonomy of values Claude expresses in practice, and an 81,000-user survey on economics of AI adoption. Methodology blends large-scale qualitative study, behavioral analysis, and policy-relevant framing.

Implications

This is Anthropic’s answer to “how do we know if alignment is working in the wild?” — the empirical alignment thread. The value taxonomy work from 700k interactions is important because it’s ground truth about what Claude actually does vs. what Constitutional AI intends. If the gap is large, it feeds back into training. The policy-relevant framing is deliberate: Anthropic uses this team’s output in regulatory conversations, EU AI Act submissions, and Congressional testimony. Watch this team’s output as leading indicators of what safety arguments Anthropic plans to make publicly.