Wellbeing Research Grants

Anthropic has launched a $5 million grant program to independently research the impact of AI models on user wellbeing. The program provides selected researchers with direct funding, model access, and technical support, with research results released as open-source projects available to everyone.

AI systems have now expanded beyond work, learning, and problem-solving for many people to include emotional support and conversational partnership. However, clear industry standards for how models should behave when users seek emotional support or experience mental health crises are still lacking.

Wellbeing assessment is a highly complex area that requires considering long conversational contexts and users' previous behavioral patterns, rather than simply checking for correct answers. For example, balanced dietary advice could be harmful to a user with a history of eating disorders. Anthropic encourages the development of independent assessments and benchmarks involving experts from various fields, including clinical specialists, psychologists, and methodologists, to capture these nuanced contexts.

This program supports research that meets the following requirements for rigorous wellbeing evaluation:

• Clearly define what is being measured (pass/fail criteria)
• Involve clinical and domain experts in design and validation
• Assess risks of both overcompliance and overrefusal
• Reflect actual user usage patterns (multi-turn conversations, context shifts)
• Validate graders against real experts

Applications are accepted until September 21, and candidates selected for full proposal submission will be notified by October 5.