Anthropic is launching a $5 million grant program to fund independent research into how AI impacts users' wellbeing. The initiative provides direct funding, access to models, and technical support to grantees building open-source evaluations that help the industry measure model effects on user welfare.
- Grantees work fully independently and publish open-source projects for developer use.
- Evaluations must clearly state what they measure and involve clinical experts in design and validation.
- Programs must test both precautions and harms, reflecting multi-turn conversations where risk escalates.
- Applications are due by September 21, with full proposal notifications by October 5.
By funding independent benchmarks, Anthropic aims to invite clinicians, psychologists, and methodologists to help develop rigorous standards for AI safety in emotional support contexts.