User Wellbeing Evaluation Grants

Anthropic is offering User Wellbeing Evaluation Grants to support independent researchers developing open-source evaluations that measure how artificial intelligence affects user wellbeing, including the mental, emotional, cognitive, and social health of people who use AI systems. Anthropic states that it invests heavily in this area internally and is providing external funding to help independent researchers conduct related work. The program expects to provide $5 million in total direct funding, with individual grants typically ranging from $500,000 to $1.5 million. In addition to direct funding, funded teams may receive access to Anthropic models through a research API or credits and lightweight technical support based on project needs. The grants are intended to fund open-source evaluations that test model behaviors, benchmarks that compare behaviors across models, and measurement methodologies that assess the broader effects of AI on user wellbeing. Funded evaluations must clearly define what they measure, validate their methods against expert human judgment, produce consistent results, and be usable by relevant model developers. Example research areas include detecting harmful dependence or emotional attachment, measuring whether model responses maximize continued engagement rather than user interests, and assessing whether models reinforce distorted thinking or negative self-talk. These examples are not exhaustive. Strong applicants are expected to combine relevant subject matter expertise, such as expertise in mental health or technology's effects on wellbeing, with deep technical understanding of AI behavior and methods for measuring it. All evaluations funded through the program must be released as open-source resources that any developer can use. Anthropic will not veto, embargo, or approve research findings. Before publication, researchers may voluntarily opt into a time-boxed factual accuracy review limited to technical claims concerning Anthropic models. Grantees are required to disclose funding received from Anthropic in all outputs. Anthropic will provide only lightweight and occasional technical support, and states that this support will not compromise the independence of the funded work. No cost-sharing or matching contribution is stated as a requirement. Eligible applicants include academic institutions, independent research organizations, and nonprofits. Individual researchers cannot apply independently and instead must submit through an institution capable of administering the grant. Applicants may be based in any country or region where Claude is available. Teams are permitted to submit more than one research idea. The source does not state additional organizational size, operating-history, faith-based, fiscal-sponsorship, or tax-status requirements. The application process has two stages. The first stage is an expression of interest containing a short description of the research idea, the proposed approach, information about the team, and a rough budget. The expression of interest is due September 21, 2026. Anthropic will select teams from this stage and contact selected applicants by October 5, 2026. Invited teams must then submit a full proposal of three to five pages containing the detailed proposed methodology, validation and grading plans, milestones, a full budget, and an ethics or IRB plan where relevant. The full proposal deadline is November 5, 2026. Applicants are instructed to review Anthropic's guidance on building a wellbeing evaluation or benchmark before applying. Anthropic will assess both expressions of interest and full proposals based on scientific rigor, feasibility, the capabilities of the applicant team, the open-source release plan, and the project's likely contribution to the field. The RFP does not provide a named program contact, email address, telephone number, application opening date, award notification date beyond the October 5 invitation notice for the second stage, project performance period, or a recurring application schedule. As of August 26, 2026, the September 21 expression-of-interest deadline and November 5 full-proposal deadline are upcoming. No annual or other recurring cycle is stated.

Demonstrate scientific rigor and feasibility; establish the team's relevant subject matter and technical capabilities; provide a strong open-source release plan; explain the project's likely contribution to the field; clearly define what the evaluation measures; validate methods against expert human judgment; design for consistent results and usability by relevant model developers.