Fund Mental Health Research with AI: OpenAI AI and Mental Health Grants 2026 (Grant Funding $5K–$100K per Project, $2M Total)

Fund Mental Health Research with AI: OpenAI AI and Mental Health Grants 2026 (Grant Funding $5K–$100K per Project, $2M Total)

OpenAI’s 2026 AI and Mental Health Grant Program is a real funding opportunity for people doing research at the intersection of artificial intelligence and mental health safety. The official call is closed now, but it is worth studying because the design of the program gives a clear model for this kind of applied safety research: practical questions, measurable outputs, and explicit boundaries.

If you are deciding whether to spend time on this opportunity, read this page as a readiness guide, not just an announcement summary. It explains what is clearly confirmed, what is not, and how to avoid wasting effort.

At-a-glance snapshot

What this opportunity is, in plain language

This is a grant program designed to fund independent research, not product development, and not a pipeline for commercial partnerships. OpenAI’s own wording is that it is from its safety systems organization and focused on research around AI and mental health, with the goal of improving understanding of both potential benefits and risks.

The practical meaning is straightforward:

• The funding is for a study, method, evaluation, or research artifact.
• The work should produce evidence and usable materials, not just a concept.
• The program is interested in how AI systems can be safer and more appropriate in mental-health-relevant contexts.

This is not a general “mental health AI startup” grant and it is not positioned as a clinical treatment grant. It is specifically about informing safety and policy-ready understanding through evidence.

Why this grant is easy to evaluate if you are a prospective applicant

Many calls are hard to evaluate because they mix vague goals with broad eligibility. This one is relatively assessable because several constraints are directly stated. You can test your own idea quickly against these anchors.

You should strongly consider applying when:

• You have a research question that is concrete, testable, and tied to AI behavior in a mental health context.
• You can clearly define what counts as success in a short to medium timeline.
• Your outputs can be described as reusable by other researchers or directly usable by safety teams.
• You have, or can realistically secure, ethics and participant safety structures.

You should usually not invest in a proposal here if:

• Your proposal is mostly a product roadmap or feature launch.
• The core outcome is a commercial app with no independent research claim.
• You cannot define measurable outcomes.
• You are unsure about handling sensitive data, participant safety, and review oversight.

What the official documents do and do not say

The program page gives reliable details in these categories:

• Open period and deadline.
• Review cadence and target notification window.
• Contact email and no-unselected-status-updates policy.
• Candidate profile requirements and for-profit prioritization note.
• Budget range and allowability language.
• Topic examples and example deliverable types.

What is not fully specified in the public text is important too. For example, it does not include a fully expanded application checklist, exact scoring rubric, required sample format, or reviewer criteria beyond topic fit and practical outputs. So, while this page can explain readiness, you should build your application around what is explicitly required and avoid guessing at hidden rules.

What this program likely supports best

The official page lists example research areas and stresses that the list is not exhaustive. This matters: your topic can sit outside those bullets if it still matches the same intent.

The listed examples can be grouped into five practical themes:

1. Language and interpretation. They explicitly seek work on distress, delusion, and language variation across cultures and languages, including where AI detection or interpretation can fail.
2. Human experience and harm perception. The page includes perspectives from people with lived experience and asks what feels safe, supportive, or harmful in AI conversations.
3. Clinical context and usage patterns. It includes research on how practitioners use AI tools, where systems work, and where harms or limitations appear.
4. Safety behavior design. It asks for work on pro-social behaviors, tone framing, youth safety interaction style, and stigma in recommendations.
5. Evaluation artifacts. It includes multimodal distress indicators (e.g., body image distress), annotated data ideas, and compassionate support patterns for grief.

These examples are strong signals of how they expect “fit” to be read: practical, behavior-focused, and tied to safety-relevant outcomes.

What the program explicitly expects to receive from funded projects

The call says these are illustrative, not exhaustive, deliverables. In practical terms, these outputs are valuable because they are inspectable:

• Research papers with evidence.
• Behavior taxonomies that can be iterated on.
• Culturally or linguistically diverse datasets (with ethical handling).
• Prototype interaction flows for context-aware support responses.

For your own planning, treat every output as “reusable evidence,” not “internal notes.” A winning proposal should indicate where a reviewer can look and verify that it improves understanding or decision-making.

Who should apply: a practical fit model

Use this as a hard filter before spending weeks writing.

Core eligibility (stated)

The application criteria listed on the official page are:

• Must be 18 years or older.
• Affiliated with a research institution or organization, and/or have significant mental health experience.
• Non-profit, research-oriented projects are prioritized; for-profit initiatives are not prioritized.
• Eligible projects should be in the budget band $5,000 to $100,000.

Who fits especially well

You are likely a strong match when at least three of the following are true:

• You can define a narrow question where AI behavior can be observed and measured.
• You can show how your work links technical AI behavior with lived mental health experience.
• You have access to the data, participants, or collaborators needed to complete research safely.
• Your proposal can produce an evaluation mechanism or actionable output for safety teams.
• You can show institutional support for supervision, ethics review, and responsible handling of sensitive materials.

Who should defer this opportunity

Do not force a fit if you are:

• Seeking unrestricted R&D funding for product development.
• Working without any documented data/privacy safeguards.
• Building a concept that depends on unverifiable claims about impact.
• Aiming for a grant that primarily funds commercialization.

How the application period worked (and what happened)

The official grant page states:

• Submissions open Dec 1, 2025 and closed Dec 19, 2025 at 11:59 PM PST.
• Reviews were rolling by a panel of internal researchers and experts.
• Selected proposals were to be notified on or before Jan 15, 2026.
• Jan 28, 2026 brought an update that applications were closed and over 1,000 entries were received.

Because the cycle is closed, there is no action needed now in this round, but the way this was run still tells applicants how to prepare for similar rounds: you need to submit a complete, coherent package and cannot assume status updates if rejected.

Eligibility, prioritization, and interpretation notes

These three points are often misunderstood:

1. Age is a hard gate. If you are under 18, you are out.
2. For-profit is not impossible, but deprioritized. The wording is explicit: this call was seeking to fund research rather than for-profit initiatives.
3. Being affiliated is flexible, but not vague. “Affiliated” can be institution-based or role-based with significant mental health experience. This broad line helps independent researchers, but still expects demonstrated capability.

The safest way to present this in your application materials is to answer: How is this project research-first? and How does this improve safety understanding?

Budget guidance you can actually use

The official allowed-cost language is broad: reasonable and necessary direct and indirect costs consistent with institutional policy.

That means your budget should be credible only if each expense is tied to execution.

When preparing a budget narrative, keep this structure:

• Person-hours by role (PI, technical lead, annotator, clinician/advisory roles).
• Participant related costs.
• Data, storage, compliance, and security-related costs.
• Analysis and validation tools directly tied to methods.
• Indirect costs if your institution permits them under policy.

Avoid budget lines that look inflated or disconnected from measurable milestones.

How to decide if this is worth your time

Use this practical self-check before committing your application cycle.

Readiness scorecard

Score each area 0-2:

• Question clarity (0 = broad and unclear, 2 = narrow, testable).
• Evidence plan (0 = assertions only, 2 = clear metrics and analysis).
• Safety and ethics design (0 = not yet, 2 = explicit workflow and escalation).
• Funding realism (0 = generic, unrealistic budget, 2 = justified line-by-line).
• Output value (0 = vague deliverable, 2 = concrete reusable artifact).

If you score less than 7, you are likely better served by pre-work before submitting. If you score 8-10, you are in a much stronger position.

Time-estimation test

If you cannot complete a draft package in under 10 focused working days, you may still be able to submit, but you likely need stronger planning. This is not a requirement from OpenAI; it is a practical threshold to prevent rushed submissions.

Fit test for teams

Each team member should be mapped to one of three buckets:

• Safety and study design.
• Domain expertise.
• Execution (recruitment, data, analysis, implementation).

A good fit has at least one strong link in each bucket.

Required materials: official vs practical

The official pages list key content expectations, but not every submission field. This is an important distinction.

Explicitly referenced in official material

• Project concept and research direction.
• Research outcomes/deliverables.
• Budget scope within $5,000-$100,000.
• Contact email for questions.
• Eligibility and submission timing details.

Practical materials you should prepare anyway

Even when not listed in exact detail, these normally improve review readiness:

• A concise plain-language summary for non-specialist reviewers.
• A method section with measurable outputs and evaluation criteria.
• A safety and participant-protection plan.
• Institutional support confirmation where required.
• A one-page budget rationale.
• Team roles and timeline.
• A dissemination plan for how outputs become reusable.

Because official review notes can only include what is present in the official portal and reviewer criteria, avoid adding extra bells and whistles that dilute the core research claim.

Common application mistakes that cost candidates time and credibility

These are frequent issues in this type of program:

1. Using a broad “AI for mental health” pitch instead of a testable research question.
2. Ignoring the distinction between a safety/research fund and a product fund.
3. Leaving outputs undefined or treating the deliverable as “to be decided later.”
4. Omitting clear safety escalation and participant protection details.
5. Submitting high-cost budgets with weak activity linkage.
6. Using generic support letters without concrete commitments.
7. Trying to match every example bullet exactly instead of building a tightly focused project.
8. Assuming status updates are available for rejected applications.

If you are applying to another cycle, address each mistake with a corresponding control before submission.

Detailed preparation plan (reusable after this cycle closes)

The following sequence is built from what OpenAI explicitly requests and what strong reviewers need to evaluate quickly.

Step 1: Define one sentence and one metric

Write one sentence that says exactly what research question you are answering. Write one metric that proves progress.

Example sentence pattern:

“This project tests whether [intervention or method] improves [AI safety outcome] for [specific user group or context], measured by [metric].”

If you cannot complete this in 2–3 lines, narrow further before writing the full proposal.

Step 2: Choose one primary deliverable

Pick one primary output that is objectively reviewable: a paper, rubric, taxonomy, dataset, or prototype interaction flow. Secondary outputs can exist, but one primary artifact must be clear.

Step 3: Draft a safety-first method section

Include:

• Data source and consent approach.
• Inclusion of expert or lived-experience input.
• Evaluation method and limitations.
• Any risk controls for participant distress.

Step 4: Align budget to milestones

Write a table that ties each budget line to a milestone date and expected deliverable.

Step 5: Build the review-ready summary

Create a one-page non-technical summary, then a technical appendix. If both sections disagree on claims or scope, revise.

Step 6: Pre-submission sanity checks

• Confirm eligibility at top.
• Confirm no prohibited status assumptions.
• Confirm you are applying before the known deadline.
• Confirm all required fields are present in the actual portal interface at submission time.

This plan is written to help teams avoid the most expensive errors in short review windows.

FAQ for this specific opportunity

Is OpenAI Group PBC really funding this?

Yes. The official page says the grants are funded and administered by OpenAI Group PBC and clarifies they are separate from People-First AI Fund and OpenAI Foundation initiatives.

Is affiliation required to apply?

The published criteria say applicants should be 18+ and affiliated with a research institution or organization, and/or have significant mental health experience.

Can a for-profit organization apply?

The program states that it is seeking research funding, and for-profit initiatives are not prioritized. That means these applications may still appear but are less aligned with stated intent.

Do they provide status updates?

No. The official communication says OpenAI will not provide status updates for applications that are not selected.

Did OpenAI select applicants in order of quality only?

The program states rolling review by a panel and later announces that all funded proposals were notified. It does not publish a public ranking rubric.

What if a future version is announced?

This is usually the most practical answer:

• Reuse your draft, but update it to match the new cycle’s language.
• Watch for a new opening date and new deadline.
• Keep your methods and ethics plan reusable.

What to do next right now

Because this cycle is closed, your best near-term action is structured readiness work rather than repeated applications to a closed form.

• Keep a short one-page concept note and budget skeleton ready.
• Build a reusable safety checklist for any future funding call.
• Track OpenAI research announcements for refreshed rounds.
• Prepare a contacts and collaborators page so you can execute quickly when a new call opens.

If you are already planning a new cycle application, do not wait for the first-day announcement to begin work. The gap between “idea formation” and “submission-ready” is usually where most teams lose quality.

Official links

• Program page: https://grants.openai.com/prog/openais_ai_and_mental_health_grant_program/
• Announcement and context: https://openai.com/index/ai-mental-health-research-grants/
• Program contact: [email protected]

Final takeaway

This opportunity is best for people who can do one thing well: deliver solid, safety-relevant evidence around AI and mental health. If you have a broad idea, narrow it. If you have a vague method, tighten it. If you can only say “we’ll figure it out,” postpone the application and improve your design first. The call is closed for this cycle, but the criteria and expectations are clear enough to use as a template for your next round.

If you are trying to decide whether to invest your time in this opportunity, start here: this is a real call for research projects where OpenAI wants evidence, not product ideas. The program funds independent research at the AI and mental health boundary, with the stated purpose of improving understanding of both risk and benefit, and informing safer AI systems.

The public material says this is a 2026 call with a program budget of up to $2,000,000, and each project grant can range from $5,000 to $100,000. Submissions were open Dec 1, 2025 and the published deadline was Dec 19, 2025 at 11:59 PM (PST). Applicants were reviewed on a rolling basis, with selected proposals notified on or before Jan 15, 2026. OpenAI also posted an update on Jan 28, 2026 that applications are closed for that cycle.

At a glance

What this grant is for in plain English

This is a safety-oriented research funding program, not a grant for startup scaling or commercial deployment. OpenAI created it to learn from independent researchers about how AI systems handle mental health contexts and where harm mitigation should improve.

Think of three things this call is most likely to support:

1. Evidence-generation around AI behavior in emotionally sensitive interactions.
2. Practical frameworks, datasets, or benchmarks that can improve model responses and evaluation.
3. Research designed to be reusable by others in the wider AI and mental health community.

The page does not promise project-level mentorship, guaranteed implementation, or status updates for applications not selected. If you apply for clarity on process, use the listed contact email.

Why this call can be worth your time

Many opportunities ask for many things. This one is easier to assess because several points are explicit and actionable.

It is likely worth your time if:

• You can define a research question around mental health and AI behavior in concrete terms.
• Your team can run a study within a realistic budget.
• You can produce evidence-oriented outputs that are shareable and auditable.
• You are prepared to follow human-subjects and data-governance standards.

It is usually not worth your time if:

• You are only looking for a commercial program.
• Your concept cannot be clearly tested in a defined study.
• You are not comfortable handling participant safety, consent, and privacy workflows.
• Your proposal relies on generic statements without measurable outcomes.

What the program explicitly funds

The official text frames this as a call for proposals that advance independent research on AI and mental health. Examples include, but are not limited to, research on:

• Cross-cultural and cross-language variation in distress language and how AI systems detect or mis-handle it.
• Perspectives from people with lived mental health experience on what feels supportive or harmful in AI conversations.
• Real-world use of AI by mental healthcare providers, including strengths and safety risks.
• AI practices that could reduce harm and increase pro-social outcomes.
• Robustness of safeguards when language is vernacular, slang-heavy, or from low-resource settings.
• Age-appropriate AI response style and framing for younger users.
• Stigma patterns in model recommendations and interaction style.
• AI interpretation of visual indicators linked to body image distress and eating disorders, where ethical dataset and annotation work can support improved evaluation.
• Compassionate support patterns for grief and emotional processing, including tone style guidelines and exemplar interactions.

The page is very clear that this list is not exhaustive, so your topic can be adjacent as long as the project fits the same practical safety and mental-health objective.

Outputs that are likely to score well

OpenAI gave examples of deliverables that can work. In practical terms, your project will be strongest when it can produce outputs that answer the question “what should someone do with this result right now?”.

Expected output types include:

• Research papers.
• Behavioral taxonomies in sensitive contexts.
• Culturally or linguistically diverse datasets (with ethical constraints).
• Prototype interaction flows for context-aware conversational patterns.

If your work only produces a vague concept without concrete deliverables, it will likely struggle. If it produces at least one usable artifact and a method that can be followed, it stands a better chance.

Eligibility checklist

The verified eligibility bullets are short, but you should treat them as hard gates:

1. You are at least 18 years old.
2. You are affiliated with an institution or organization, or you have significant mental health-related experience.
3. You can justify that your proposal is research-oriented rather than a for-profit product launch.

The language implies non-profit and academic work is preferred in practice. It does not say only universities can apply, but it does say for-profit initiatives are not prioritized.

Budget, costs, and what you can include

The page says allowable costs include reasonable and necessary direct and indirect costs consistent with institutional policy. Practical budget categories that usually pass review are:

• Research personnel effort.
• Participant compensation and recruitment costs.
• Annotation, translation, or evaluation tooling if core to your method.
• Secure computation and storage.
• Core privacy and security controls for sensitive data.
• Institutional overhead where allowed.

A good way to avoid budget failure is to align each line with a named activity and timeline. Do not include costs that are not tied to a specific step in your study.

How to prepare an application that is review-ready

Use this workflow, especially if you are preparing for a future cycle:

1. Write a one-page plain-language version first. If you cannot explain the idea in plain language, your technical detail is probably too dense.
2. Write a methods section with explicit outcomes and how you will measure them.
3. Define participant pathway and safety safeguards before filling your data section.
4. Draft a reproducibility plan, including when and how data/code or taxonomies will be shared.
5. Build a budget that maps exactly to your timeline.
6. Collect support letters that commit resources (access to participants, clinical oversight, computing, or language expertise).
7. Perform one external read-through by someone outside your discipline.

If you are submitting as a team, confirm each member’s role and deliverable ownership before final submission.

Decision quality: should your proposal be accepted?

A useful way to self-score your draft:

• Score 1 to 5: How specific is your research question?
• Score 1 to 5: Are outcomes measurable?
• Score 1 to 5: Is participant safety addressed in writing?
• Score 1 to 5: Is the budget justified and realistic?
• Score 1 to 5: Are outputs concrete and reusable?

If the total is below 20 out of 25, this application is likely to be weak in at least one review-critical area.

The 2026 call rewarded projects with clear outputs, not only strong ideas. Focus on output quality and methodological credibility, not novelty headlines.

To further increase your odds, read your own draft against a reviewer lens:

• Does the proposal explain why this question matters for safety specifically, rather than for AI generally?
• Are the methods sufficient to test your question, or do they depend on unsupported assumptions?
• Can a reviewer from another institution understand why your sampling and analysis choices are correct?
• Is the team composition proportional to the question, including mental health expertise and technical execution?
• Are deliverables concrete enough that OpenAI can see how the work can be reused?

If you can answer all five clearly, you are much closer to a fundable submission than a proposal that relies on topic relevance alone.

Another practical point: because OpenAI noted rolling review, applications were designed to be considered once complete rather than as perfect, polished documents in a single weekend sprint. This means you can improve quality by iterating, but you should not delay to the point where operational risk increases near deadline.

At-a-glance decision questions before investing 20+ hours

Use these questions before you start building the full package:

• Can I submit all required sections with confidence in under two hours, or do I keep rewriting the same section?
• Do we have clear access to the population, platform context, or data source we need?
• Does my proposal avoid jargon without diluting technical correctness?
• Have we written one paragraph specifically explaining why this research is not a product feature launch?
• Do we have ethics readiness and escalation handling documented?
• Can external readers understand the potential impact in one page?

If you are not confident on the above, do those tasks first, then proceed.

Required materials list

Based on the call text and what similar OpenAI program pages require, the practical list should include:

• Project narrative with aims, methods, expected results, and timeline.
• Budget with direct and indirect cost breakdown.
• CVs or biosketches of key team members.
• Letters of support that define concrete commitments.
• Ethics or data protection plan that covers privacy, storage, and review.
• Dissemination and sharing plan for outputs.

For mental-health-adjacent work, reviewers usually care most about safety, consent design, and what happens if distress appears in interactions.

Common mistakes that quietly sink applications

1. Building a broad proposal with too many goals.
2. Skipping safety escalation and escalation timing.
3. Writing objectives without measurable outcomes.
4. Ignoring indirect costs needed to run compliant research.
5. Using generic support letters.
6. Waiting until the end to verify deadlines and portal details.
7. Mixing product roadmap language into a research proposal.

The fix is simple: narrow, measure, justify, and submit cleanly.

Practical timeline for applicants when a future wave opens

A robust internal schedule if OpenAI runs another wave:

• Week 1 to 2: Define research question, eligibility proof, and target output type.
• Week 3: Draft study design and participant flow.
• Week 4: Draft budget and ethical safeguards.
• Week 5: Collect collaboration commitments and review support materials.
• Week 6: Internal review for clarity and feasibility.
• Week 7: Finalize final package and submit early.

Even with rolling review, submitting early avoids avoidable submission-system risk and gives you time to correct errors before hard closure.

What to do if this 2026 call is already closed for you

The official update is clear: this cycle is closed. In that situation, the best next step is not to force an already closed round.

Use this instead:

• Read the call text once more and map your idea to the same criteria.
• Keep your one-page proposal template and improve it.
• Track OpenAI and its research pages for new calls.
• If your idea is strong, prepare it in reusable form and submit when a new window opens.

FAQ

Q: Can non-academic professionals apply? A: Yes, if they have significant experience in mental health and can show research readiness. The eligibility wording uses affiliation OR significant experience.

Q: Is this for a single institution only? A: The call wording does not restrict by one institution. It is tied to institutional affiliation or experience and global submission patterns described by OpenAI.

Q: Do I get updates if not selected? A: OpenAI states no individual status updates for applicants who are not selected.

Q: Can for-profit organizations apply? A: For-profit initiatives are not prioritized for this program.

Q: What should I put in a proposal if I apply later? A: Emphasize measurable outcomes, safety workflows, participant protection, and concrete reusable outputs.

Official links and next actions

• Official application page: https://grants.openai.com/prog/openais_ai_and_mental_health_grant_program/
• Program explanation and announcement: https://openai.com/index/ai-mental-health-research-grants/
• Program contact: [email protected]

If you are reading this before a new cycle starts, prepare your project now so you can submit in the first week when reopening is announced.