Humanity AI's $10 Million Open Call Is Finally Live — and It Rules Out the Three Things Most Nonprofits Would Propose

Since May 2026, when the Humanity AI collaborative announced its first $18 million in grants and promised an open call "this summer," nonprofits working on artificial intelligence have been waiting on two questions: how much, and who qualifies.

The answers arrived September 10, 2026. The open call is $10 million, applications are due October 21, 2026 at 4:59 p.m. Pacific (7:59 p.m. Eastern), and the eligibility rules are considerably narrower than the collaborative's public framing suggested.

The narrowing is the story. Humanity AI has published an explicit out-of-scope list, and it rules out three of the most common things a nonprofit would propose to an AI funder.

The three things it will not fund

Directly from the open call guidelines, out of scope:

• AI literacy training
• Promoting AI adoption
• Research-only projects

That list eliminates an enormous share of what the nonprofit sector currently does under the heading of "AI work." A community organization that wants to run workshops teaching residents how to use AI tools is out. A workforce nonprofit that wants to help small businesses adopt AI is out. A university center that wants to study AI's labor effects and publish findings is out.

This is not an oversight. Humanity AI's three eligible approaches describe something different from education or study:

1. Shaping AI governance and use through advocacy
2. Building AI on community terms — technical infrastructure, datasets, audits
3. Challenging misaligned AI deployments

Those are power-oriented verbs. Shaping, building, challenging. The collaborative's press language is about supporting "communities' collective power to shape how AI is built, governed and used," and the funding criteria enforce it. If your project's theory of change ends at "participants will understand AI better," it does not fit.

The three issue areas

Proposals must sit in one of three domains:

• Education — shaping AI so it benefits students and educators, with student- and educator-led initiatives prioritized
• Labor and economy — enhancing work rather than replacing it, with worker-focused efforts prioritized
• Humanities, arts, and culture — protecting creators while enhancing creativity, with artist- and creator-led work prioritized

The pattern across all three is the same: the constituency affected should be leading, not being served. That maps directly onto one of the five review criteria, Vision for Community Leadership, which asks whether community members define the goals and shape the approach.

Humanity AI also states it is actively inviting applicants not traditionally involved in AI policy. Combined with the community-leadership criterion, that is a fairly clear signal that an established AI policy shop submitting a standard advocacy proposal is not the profile being recruited.

Three tiers, and what each signals

Grants run $75,000 to $1,000,000 over up to three years, structured in tiers:

The number of awards is not predetermined — it varies with application quality. That is an unusual and meaningful disclosure. In a fixed-award competition, you can reason about your odds from the ratio of awards to applicants. Here you cannot, and the tier you choose changes the arithmetic materially: $10 million is 100 Tier 1 grants at the ceiling, or 10 Tier 3 grants at the ceiling, or any mix.

One review criterion explicitly ties the two together. Clarity and Feasibility asks whether the plan is "specific, realistic, and matching the funding tier." A coalition-scale vision submitted at Tier 1 reads as underfunded; a single-organization pilot submitted at Tier 3 reads as unserious. Pick the tier that matches the work, then size the work honestly.

Grants are disbursed at the beginning when feasible — front-loaded rather than reimbursed on milestones. For a small organization, that cash-flow structure is worth real money and worth noting in your planning.

Who can apply

Eligible applicants:

• U.S.-based 501(c)(3) organizations
• Collaborations with a 501(c)(3) lead applicant
• Individuals, through fiscal sponsorship or collaboration
• Tribal governments treated as States under IRC Section 7871
• For-profits may participate but cannot lead

All proposals must address U.S.-based communities, though non-U.S. entities may participate within collaborations.

Two restrictions catch people:

Organizations that received a 2026 Humanity AI grant cannot apply as sole applicants — but may join collaborations. That closes the door on the roughly dozen inaugural grantees named earlier this year while leaving them available as partners, which is arguably the more useful role for a first-time applicant to recruit them into.

Regranting cannot be the sole purpose of a proposal. An intermediary whose entire model is redistributing funds needs a programmatic spine beyond the redistribution.

Application volume limits are firm: one individual application and one collaboration application per applicant. Separate centers or departments within the same organization may apply separately if the proposals are genuinely distinct.

The open-weight preference

One technical criterion deserves attention because it will disqualify otherwise strong proposals from consideration at the margin:

If AI models are central to project methodology, priority goes to work that integrates open-weight and/or open-source models over closed-weight models.

Routine AI use — drafting, research assistance — is explicitly permitted with no preference attached. But if models are core to your method, building on proprietary APIs puts you behind proposals built on open weights.

This is philosophically consistent with the funder set. Mozilla Foundation is a founding partner, and the "building AI on community terms" approach is about infrastructure communities can actually control. A community auditing tool that depends on a commercial API the vendor can revoke or reprice is, by the collaborative's own logic, not on community terms.

If your project's method is model-dependent, budget the engineering time to work with open-weight models, and say so explicitly in the proposal.

Who is funding this

Ten foundations anchor the collaborative: the Doris Duke Foundation, Ford Foundation, Lumina Foundation, Kapor Foundation, John D. and Catherine T. MacArthur Foundation, Mellon Foundation, Mozilla Foundation, Omidyar Network, David and Lucile Packard Foundation, and Siegel Family Endowment.

The initiative is a five-year, $500 million commitment. This $10 million open call is a slice of it — and for most nonprofits, the only slice they can reach by application. The MacArthur Foundation does not accept unsolicited proposals. Neither, as a general matter, do Gates or Bloomberg. An open call from a pooled fund of this composition is a genuinely rare access point.

The process and the calendar

Applications run through Submittable, administered by Lever for Change — the MacArthur-affiliated intermediary that runs 100&Change and similar competitions. The review is two-phase:

Phase 1 — Lever for Change screens for eligibility, then evaluates against five criteria: Vision for Proposal (clear, compelling, beyond existing work); Vision for Community Leadership; Clarity and Feasibility; Potential for Impact (advancing community power or generating field learning); and Credibility and Experience.

Phase 2 — Finalists submit budgets and measurement plans; some participate in interviews.

Note the gap: from submission to money is roughly six months. This is not bridge funding for a current shortfall.

Also note what is not available: pre-submission feedback, which the collaborative says it cannot offer given anticipated volume. There is a recorded webinar, and there is the written guidance. That is the extent of the coaching.

The strategic read

Six weeks is a short runway for a $1 million proposal and a comfortable one for a $100,000 pilot. That asymmetry is worth taking seriously.

If you are assembling a Tier 3 coalition from scratch right now, you are behind — multi-partner budgets, shared governance, and letters of alignment take longer than the calendar allows, and Clarity and Feasibility is a scored criterion. If you have an existing coalition with a live agenda, the tier fits.

If you are a smaller organization, Tier 1 is not a consolation prize here. The number of awards is undetermined, the tier is explicitly designated for "pilots, early initiatives, focused proposals," and the funder is openly recruiting outside the established AI policy field. A sharp, narrow, community-led Tier 1 proposal from a labor or arts organization that has never touched AI policy is close to the archetype described in the guidelines.

The failure mode to avoid is the one the out-of-scope list was written to prevent: reframing an existing training program, adoption initiative, or research agenda in the language of community power. The five criteria ask what the community will be able to do that it cannot do now. Proposals that answer "know more" are answering a different question.

For background on the collaborative's first grant round and its earlier signaling, see our coverage of the $18 million inaugural cohort and the $500 million five-year commitment.