Anthropic puts $5M behind AI wellbeing tests

Anthropic launched a $5 million grant program for independent, open-source research into how AI affects user wellbeing, with applications due September 21, according to Anthropic’s grant announcement. The company is asking applicants to test both harms and over-refusal across multi-turn conversations, involve subject-matter experts, and validate automated graders against experts. Teams shipping conversational products should treat long-horizon wellbeing evaluation as a distinct measurement budget, extending the archive’s analysis of ChatGPT’s health-data perimeter.