The programme will support independent, open-source evaluations of how AI affects user wellbeing, with direct funding, model access and technical support.
What you need to know
- $5 million programme with direct funding, model access and technical support
- Applications close on 21 September 2026
- Projects should test realistic multi-turn interactions and validate automated graders with humans
What Anthropic wants to fund
The call prioritises rigorous evaluations of both harmful model behaviour and excessive refusal. It asks for realistic conversations, suitable clinical or subject-matter expertise and evidence that automated scoring reflects expert human judgement.
Open work is part of the bargain
Supported research is expected to be open-source. That makes the programme relevant to teams able to produce reusable measures, datasets or evaluation methods rather than private product studies with no public evidence.
The application window is short
Initial applications close on 21 September. Anthropic says selected applicants will be invited to submit full proposals from 5 October. Applicants should define the harm or wellbeing outcome, test population, human-validation plan and deliverables before submitting.
HUBAI VIEWStrong proposals need realistic multi-turn tests, relevant subject expertise and human validation—not another synthetic benchmark with no connection to lived outcomes.
Buyer decision signal: Funding deadline · 21 September
What to verify next
1Define a measurable wellbeing outcome
2Include relevant domain expertise
3Test harms and excessive refusal
4Plan human validation of graders
5Submit before 21 September 2026
Read the evidence
Capabilities, availability and prices can change. HubAI keeps analysis separate from the underlying official material.
01Anthropic: Wellbeing research grantsOpen source ↗
Safety
Business
Products
Code AI