Anthropic AI Wellbeing Research Grants: Apply by September 21

Anthropic has launched a $5 million grant program for independent research that improves how AI systems are evaluated for their effects on human wellbeing.

The program is not a general-purpose research fund and it is not primarily asking for another chatbot-satisfaction survey. Anthropic is looking for open-source evaluations and benchmarks that can test difficult, context-dependent interactions—for example, when a user increasingly relies on an AI companion, discusses disordered eating or moves toward a mental-health crisis over several turns.

Expressions of interest are due September 21, 2026. Selected applicants will be invited to submit full proposals and are expected to be notified by October 5.

What the program offers

Anthropic says the overall program will provide:

• Direct research funding from a total pool of $5 million.
• Access to Anthropic models.
• Technical support for selected grantees.
• A pathway for publishing evaluations and benchmarks as open-source work that other developers can use.

The public announcement does not state a standard award amount per project, a fixed project duration or a guaranteed number of grants. Applicants should avoid building a budget around an assumed award size until Anthropic provides proposal-stage guidance.

What kind of research Anthropic wants

Anthropic’s published guidance emphasizes evaluations that:

1. Clearly define what they are measuring and why a pass or failure matters.
2. Include clinical or other relevant subject-matter experts in design and validation.
3. Test both under-reaction and over-reaction—for example, failing to respond safely versus unnecessarily refusing benign support.
4. Reflect how people actually use conversational AI, including multi-turn interactions where risk and context change over time.
5. Validate automated graders against qualified human experts.

This makes the opportunity a particularly strong fit for interdisciplinary teams. A technically sophisticated benchmark without clinical validity may be too narrow. A clinically thoughtful framework without a reproducible technical evaluation may also be incomplete.

Who should consider applying

Anthropic specifically invites expertise from fields such as clinical practice, psychology and evaluation methodology. Strong applicants may include:

• University researchers studying human-computer interaction, psychology, psychiatry, public health or AI safety.
• Clinician-researchers working on crisis response, eating disorders, dependency, loneliness or related wellbeing concerns.
• Benchmark and evaluation specialists who can build reproducible multi-turn tests.
• Nonprofit or civil-society research teams with relevant communities, ethical safeguards and technical partners.
• Interdisciplinary groups combining lived-experience input, clinical knowledge and model evaluation.

The announcement does not provide a complete geographic or institutional eligibility list. Check the official application form for restrictions that apply to your organization, location or proposed budget.

What a strong expression of interest should explain

Prepare a concise answer to five questions:

1. What wellbeing risk or protective behavior are you measuring? Avoid using “wellbeing” as an undefined umbrella term.
2. Why does the behavior require a multi-turn evaluation? Explain what would be missed by judging one isolated response.
3. Which experts and affected communities will shape the standard? Name the role of clinicians, methodologists and people with relevant lived experience.
4. How will you validate the grader? Describe how automated scoring will be compared with qualified human judgment.
5. What will become openly reusable? Identify the scenarios, rubric, code, documentation or dataset you can responsibly publish.

Also address privacy and research ethics early. Wellbeing scenarios can involve highly sensitive information. A useful benchmark does not require collecting unnecessary personal data or exposing identifiable crisis conversations.

Important questions to resolve before applying

The fund is announced by the same company whose models may be evaluated. Anthropic says grantees will work independently and publish open-source results, which is a meaningful design choice. Applicants should still clarify publication rights, disclosure expectations, data access, safety review and whether unfavorable results can be released without restriction.

Confirm which costs are eligible, whether indirect costs are permitted, who owns resulting code and datasets, and what ethics approval is required. If the work uses real conversations or participants, establish data-minimization, consent, safeguarding and crisis-response procedures before collection begins.

Applicants should also treat the $5 million figure as the total program pool, not the expected value of an individual grant.

How to apply

1. Read Anthropic’s official grant announcement and evaluation criteria.
2. Follow the application link from the official page.
3. Submit the expression of interest by September 21, 2026.
4. If selected, prepare for a full-proposal invitation; Anthropic says notifications are expected by October 5.
5. Confirm your institution’s ethics, data-protection and grant-acceptance requirements before committing sensitive research activity.

For research-quality habits, use AIViewer’s verification workflow. For the broader product context, read the ChatGPT and Claude privacy-settings guide and AI basics for models, assistants and agents.

AIViewer verdict

This is a serious, time-sensitive opportunity for teams that can connect clinical or wellbeing expertise with reproducible AI evaluation.

Apply if you can define a narrow behavior, build a realistic multi-turn test, validate it with qualified people and release something the wider field can reuse. A broad proposal about “making AI safer” is unlikely to be as useful as a precise evaluation with a defensible pass condition and a clear account of both harms and false alarms.

Do not apply merely because your work mentions AI or mental health. The announced criteria point toward methodologically rigorous evaluation infrastructure, not a general awareness project or product-development grant.

Frequently asked questions

How much funding is available?

Anthropic announced a total program pool of $5 million. It did not publish a standard amount for each selected project in the announcement.

When is the deadline?

Expressions of interest are due September 21, 2026. Anthropic says selected applicants will be notified by October 5 to submit full proposals.

Must the work be open source?

Anthropic describes the program as funding open-source evaluations and benchmarks that any developer can use. Applicants should confirm the specific licensing and publication requirements during the proposal process.

Is this only for universities?

The announcement invites clinicians, psychologists, methodologists and others but does not present the program as university-only. Check the application form for the final organizational and geographic requirements.

Does the program fund a wellbeing product or service?

The announcement focuses on evaluations and benchmarks. A proposal centered mainly on launching a consumer product, therapy chatbot or awareness campaign appears less aligned unless it produces the rigorous, reusable evaluation work the program requests.

Primary source

• Anthropic: Funding better evaluations of AI’s impact on wellbeing, published August 25, 2026 and checked August 30, 2026.

AIViewer did not review the grant agreement or full-proposal materials. The official announcement and application form remain the authority.