FuriaHubAnthropic opens a $5 million grant program for wellbeing evals
Anthropic is funding open-source evaluations of how models affect human wellbeing, with money, model access, and technical support. Applications are due September 21, 2026, and full proposals by October 5; funded work must be published as open source.
On August 25, 2026 Anthropic announced a $5 million grant program for open-source evaluations of how AI models affect wellbeing.
What we know
- The $5 million program offers funding, model access, and technical support for open-source wellbeing evaluations.
- Recipients must publish the work as open source.
- Applications are due September 21, 2026, and full proposals by October 5.
- Anthropic wants clinician-designed evals that test overcompliance and overrefusal.
- The lab said wellbeing risks can emerge over multi-turn chats, not only in a single reply.
Takeaways
- Anthropic is putting $5 million behind public wellbeing evals.
- The work has to ship as open source.
- The lab is specifically asking for clinician-designed tests of overcompliance and overrefusal.
Source: Anthropic
