ED's AI Grant Priority: Funding, No Safeguards

On 13 April 2026 the US Department of Education published a final Secretary’s Supplemental Priority and Definitions on Advancing Artificial Intelligence in Education at 91 FR 18774 (FR Doc. 2026-07087), effective 13 May 2026. It is not a statute, not a standard, and not a voluntary commitment. It is a grant priority: a clause the Secretary can drop into any Department discretionary grant competition to make AI work a scoring advantage, a bare-minimum eligibility test, or a stated interest. It is, in other words, governance exercised through appropriations — and the most interesting thing about it is what it declines to require.

Instrument: Secretary’s Supplemental Priority and Definitions (final), not a regulation of general applicability
Citation: 91 FR 18774, published 13 April 2026, FR Doc. 2026-07087
Effective: 13 May 2026
Scope: available for use in any Department discretionary grant program, whole or in component parts
Comments received on the proposal: over 300 parties (proposed version: 90 FR 34203, 21 July 2025)
Safety, privacy and parental-consent requirements added: none — expressly declined and deferred to state and local decision-makers
Already in live use: yes — as a competitive preference priority in the FY 2026 SEED competition (ALN 84.423A), applications due 1 June 2026

What a “supplemental priority” actually does

A Secretary’s supplemental priority is a reusable piece of grant-competition text. Once finalised, the Department can attach it to individual competitions without running a fresh rulemaking each time, and the notice for each competition decides how much weight it carries. The Education Department General Administrative Regulations set out three settings, and the final priority quotes all three:

Absolute priority: “we consider only applications that meet the priority (34 CFR 75.105(c)(3)).” An application that does not address AI is not reviewed at all.
Competitive preference priority: “we give competitive preference to an application by (1) awarding additional points, depending on the extent to which the application meets the priority (34 CFR 75.105(c)(2)(i)); or (2) selecting an application that meets the priority over an application of comparable merit that does not meet the priority (34 CFR 75.105(c)(2)(ii)).” This is the setting most competitions use, and the one that moves behaviour without forcing it.
Invitational priority: “we are particularly interested in applications that meet the priority. However, we do not give an application that meets the priority a preference over other applications (34 CFR 75.105(c)(1)).” Signalling only.

The Department also reserves the right to slice it up: the Secretary may use an entire priority for a program or competition, or use one or more of the priority’s component parts. So the operative question for any given competition is never “does the AI priority exist” but “which paragraph of it did this notice pick up, and at what weight.”

What the priority covers: parts (a) and (b)

Part (a) — expanding understanding of AI. Projects that build AI literacy into teaching practice, including the ability to detect AI-generated disinformation; age-appropriate AI and computer science instruction in K-12; AI and computer science in higher-education general education; AI content embedded in preservice and in-service teacher development; professional development on AI fundamentals across subject areas; dual-enrolment AI coursework; high-school AI courses and certification pathways; and dissemination and evidence-building around AI integration methods.

Part (b) — expanding appropriate and ethical use. Projects that use AI to serve gifted, talented and accelerated learners; students below grade level who need remediation; early intervention and special education services; personalised learning through adaptive technologies and data analytics; teaching and tutoring resources; teacher preparation and training; classroom and administrative efficiency; instructional resources and college/career navigation; universal design for learning; and improved programme outcomes.

Read as a list of fundable activities, part (b) is the striking half. It invites federally funded AI deployment onto exactly the populations where error costs are highest — students receiving special education services, students already behind, students being routed towards colleges and careers by an adaptive system. The phrase doing the governance work in that paragraph is “appropriate and ethical use,” and the priority does not define what would make a use inappropriate or unethical.

The definitions, and the one that carries weight

The priority supplies its own definitions, which then bind the competitions that adopt it:

Artificial intelligence: “a machine-based system that can, for a given set of human-defined objectives, make predictions, recommendations, or decisions influencing real or virtual environments.” This is the familiar US statutory formulation, and it is deliberately wide — it takes in a gradebook analytics module as readily as a frontier chatbot.
AI literacy: technical knowledge plus durable skills including “ethical reasoning, critical social inquiry, interdisciplinary problem-solving, and creativity, required to thrive in a world influenced by AI.”
Computer science: the study of computers and algorithmic processes, spanning principles, hardware and software, computational thinking, coding, machine learning and AI, with an emphasis on problem-solving.

The AI-literacy definition is the closest thing in the document to a safety concept, and it is aimed at students rather than at systems. “Ethical reasoning and critical social inquiry” is a curricular outcome to be taught; it is not a duty owed by the grantee that is deploying the tool. Nothing in the definitions creates an obligation to evaluate a tool before putting it in front of children, to disclose that an AI system is in the loop, or to keep a human decision-maker in a placement or intervention decision. For how differently the same word behaves when a framework does define it as an obligation, compare CASRAI’s breakdown of the eleven scope tests that decide who counts as “frontier AI”.

What the Department was asked for, and declined

Over 300 parties commented on the July 2025 proposal. A recurring ask was for federal conditions to travel with the money: student data-privacy requirements, safety requirements, parental notification and opt-out before AI tools are used with children, bias and discrimination testing, and evidence of efficacy before deployment.

The Department’s answer on safety is stated plainly in the final document: “The Department believes that how best to ensure safety and communicate about technology use is optimally decided at the state and local level and declines to enact requirements at the federal level.” On parental consent, it made no mandate, framing families and educators as the parties best placed to weigh benefits against student wellbeing. On bias and evaluation, its position was that existing grant requirements already apply to Department-funded AI projects and that no new ones were needed in the priority. It states a commitment to upholding student-privacy protections under existing law and to the central role of families, without converting either into a term of the grant.

What a grantee is left with, therefore, is the pre-existing floor: projects funded under this priority must comply with applicable federal education, disability and civil-rights law, plus the standing administrative requirements every Department discretionary grant carries. That floor was written for a world of paper files, staff decisions and classroom materials. None of it was drafted with an adaptive system making placement recommendations in mind, and the priority adds nothing to bridge the gap.

This is a familiar shape in US AI governance rather than an aberration. CASRAI has documented the same pattern at the human-subjects regulator: SACHRP’s AI recommendations to OHRP remain unactioned, leaving IRBs to reason from a Common Rule that never contemplated AI. The Blueprint for an AI Bill of Rights set out exactly the protections — notice, human alternatives, algorithmic discrimination protection — that commenters asked the Department to attach here, and it was non-binding too.

Why the mechanism matters more than the text

Most of the instruments in this cluster work by prohibition or disclosure: a statute that forbids, a framework that requires a published safety case, a standard a body can certify against. A grant priority works differently. It does not tell anyone what they may not do. It changes the expected value of writing AI into a proposal, and it does so at the point where institutional behaviour is actually set — the competition notice that a sponsored-programs office reads before deciding what to submit.

That makes it a quieter lever than an executive order and a faster one than rulemaking. Compare EO 14179 and EO 14365, which direct federal agencies and set policy posture but do not by themselves score anyone’s application; a supplemental priority is the instrument through which that posture reaches a university’s proposal calendar. It is also part of a series: the Department finalised a Career Pathways and Workforce Readiness priority at 91 FR 18780 on the same day, and a Promoting Patriotic Education priority followed on 22 May 2026 (FR Doc. 2026-10347), cross-referencing the AI priority published in April. Supplemental priorities are how this Department is steering, and AI is now one of the steering wheels.

The effect is not hypothetical. The FY 2026 Supporting Effective Educator Development competition (Assistance Listing Number 84.423A), announced 20 April 2026, lists Advancing Artificial Intelligence in Education among its competitive preference priorities, alongside Returning Education to the States and Career Pathways and Workforce Readiness. Eligible applicants include institutions of higher education and national non-profits with demonstrated effectiveness, and proposals were due through Grants.gov by 11:59:59 p.m. Eastern on 1 June 2026. Seven weeks after the priority was finalised, it was already scoring applications from universities.

What this means for research administration

This is the rare AI-governance instrument that lands directly in a sponsored-programs office rather than in general counsel’s inbox. Some practical consequences:

Proposal development. When a Department competition adopts the priority as a competitive preference, addressing it is worth points, and the notice — not the priority — states how many. Read the competition notice for which component parts were picked up and at what weight before drafting; the priority’s own text does not tell you.
Absolute-priority risk. The same text can be used as an absolute priority, in which case an otherwise strong proposal that does not address AI is not reviewed. Institutions tracking Department competitions should treat “which setting” as a go/no-go question, not a scoring detail.
Nobody else is imposing the safeguards. Because the Department declined to attach federal safety, privacy and parental-consent conditions, whatever protections a funded project has will be the ones the institution wrote into its own proposal or its own policy. Commitments made in a narrative to win points are still commitments the award will be administered against.
Human subjects. Projects that study students — which part (a)’s evidence-building language actively invites — remain subject to the institution’s human-subjects review under the Department’s own human-subjects regulations and the institution’s assurance. The AI priority changes nothing there, which is precisely the problem described in CASRAI’s guide on IRBs and AI above: review boards are applying a framework that predates the technology.
K-12 partners. Higher-education applicants who subaward to districts are exporting the same gap. The district, not the Department, decides what consent and safety practice applies, and that decision is now a project-design question for the prime.

The NIKOLAI angle: a priority with no safeguard slot

NIKOLAI is CASRAI’s own independent frontier-AI-safety dictionary. It is unendorsed: no lab, evaluator or regulator has reviewed it or been consulted on it, and every crosswalk row in it is a shadow mapping — CASRAI’s reading of a published document — unless the organisation concerned has filed a Mapping Declaration. Nothing below is a claim about what the Department of Education thinks.

The useful element here is Safeguard, in NIKOLAI’s N6 track on mitigations and security. NIKOLAI proposes defining a safeguard as “a technical or procedural measure intended to reduce misuse or misalignment risk, identified by type (e.g. safety training, filter, monitoring, access control), the risk domain it targets, and the deployment scope it applies to.” That definition is really a three-field form: what kind of measure, against what risk, over what scope.

Hold ED’s priority up against that form and all three fields come back blank. The priority names no measure type, identifies no risk domain, and sets no deployment scope — it funds deployment into special education, remediation and personalised learning without specifying what is being guarded against or where the guarding applies. The element page itself currently carries nine shadow-mapping rows drawn from frontier-lab frameworks, the EU AI Act, California SB 53 and evaluator methodologies; the common feature of those documents is that each has something to put in at least one of the three fields. A funding priority that says “appropriate and ethical use” and stops has nothing to map. That is an observation about an absence, not a crosswalk row, and it should not be read as one.

Frequently asked questions

Is the AI priority a regulation that schools have to follow?

No. It is a priority available for use in Department discretionary grant competitions. It binds nobody who does not apply for, or hold, a grant under a competition that adopted it. Its force is over applicants and grantees, not over schools generally.

Does it require parental consent before AI is used with students?

No. Commenters asked for parental notification and opt-out requirements; the Department declined to impose them federally and left the question to state and local decision-makers, and to families and educators.

Does it require any safety testing, bias testing or evidence of efficacy?

No. The Department’s position was that how best to ensure safety is decided at state and local level, and that existing grant requirements already apply to Department-funded AI projects. The priority itself adds no testing or evaluation condition.

What law still applies to a funded AI project?

Applicable federal education, disability and civil-rights law, and the standing administrative requirements attached to Department discretionary grants. Those obligations pre-date the priority and are unchanged by it.

Does it mandate a particular AI product or vendor?

No. The final priority names no product, vendor or platform.

How would an applicant know whether it applies to their competition?

From the competition’s own notice. The Secretary may use the whole priority or selected component parts, and as an absolute, competitive preference or invitational priority. The notice inviting applications is what states the choice.

Is it in use yet?

Yes. The FY 2026 SEED competition (ALN 84.423A), announced 20 April 2026 with a 1 June 2026 deadline, listed it as a competitive preference priority.

Sources

US Department of Education, “Final Priority and Definitions — Secretary’s Supplemental Priority and Definitions on Advancing Artificial Intelligence in Education,” 91 FR 18774 (13 April 2026), FR Doc. 2026-07087 — priority parts (a) and (b), the AI / AI literacy / computer science definitions, the 34 CFR 75.105 priority types, the response to comments on safety, privacy and parental consent, and the 13 May 2026 effective date.
Federal Register API, documents matching “Advancing Artificial Intelligence in Education” — confirms the 91 FR 18774 publication, the 21 July 2025 proposed priority at 90 FR 34203, the Career Pathways priority at 91 FR 18780, and the 22 May 2026 Patriotic Education priority (FR Doc. 2026-10347) cross-referencing the AI priority.
US Department of Education, “Notice Announcing Supporting Effective Educator Development Program Competition,” FR Doc. 2026-07672 (20 April 2026) — the AI priority in live use as a competitive preference priority, eligible applicants, and the 1 June 2026 Grants.gov deadline.

Quoted sentences are taken from the final priority as published on govinfo.gov. Where this page describes the Department’s response to comments without quotation marks, it is summarising that response rather than quoting it.

The Education Department’s AI Grant Priority: Funding AI Adoption Without Attaching Safeguards