Wellbeing Research Grants

The organization is launching a $5 million grant program to support independent research on how AI systems affect user wellbeing. Grantees will receive direct funding, access to the provider’s models, and technical assistance to develop open‑source evaluations that can be used by the broader AI community to measure model impact on wellbeing. The grants are intended for fully independent work, with all resulting evaluations published as open‑source projects.

Evaluating wellbeing poses unique challenges because it often depends on longitudinal context rather than single‑turn accuracy. For instance, signs of distress may emerge gradually, and a response that is appropriate in one scenario—such as offering fitness advice—could be harmful if the user has a history of disordered eating. The program will therefore fund the creation of benchmarks and evaluation frameworks that capture these nuances, and it will provide guidance from the provider’s Safeguards team on what constitutes a rigorous wellbeing assessment. Applications are due September 21, with full‑proposal notifications by October 5.

• Grant size: $5 million
• Access: model usage and technical support
• Output: open‑source wellbeing evaluations and benchmarks
• Deadline: Sept 21 (applications); Oct 5 (full‑proposal notice)

Why this matters

The source material states that the grant aims to produce independent, open‑source evaluations to address the current lack of standardized metrics for AI‑driven wellbeing impacts. From this, one can infer that establishing such shared benchmarks could enable more comparable safety assessments across models, potentially informing future regulatory or industry guidelines. However, the actual influence on model design or policy will depend on the quality and adoption of the funded evaluations, which remains to be seen.