DARPA And NSF Quietly Stood Up A Joint Forum For National-Security AI Research. The June 22 RFI Is The Door. The Forum Is The Building.

On June 1, 2026, the Defense Advanced Research Projects Agency and the National Science Foundation jointly announced a new program called AI Forge. The press release was short — a few hundred words on each agency's website, a parallel notice from the National Institute of Standards and Technology's Center for AI Standards and Innovation, and a single Request for Information posted on sam.gov with a closing date of June 22, 2026. There was no dollar figure attached, no list of funded principal investigators, no Notice of Funding Opportunity in the traditional sense. To a researcher skimming federal funding news at the end of last week, AI Forge looked like a tentative interagency announcement of the kind that often precedes a real solicitation by six to twelve months.

That reading is wrong. The announcement is not a precursor to a future solicitation — it is the standing-up of a permanent venue for federal AI research that will operate outside the normal grantmaking timelines of both DARPA and NSF, governed by a nonprofit entity launching this summer, and structured to fund fast-paced university research with frontier-scale compute and model access supplied by commercial AI companies under arrangements that the press release does not detail. The forum model is the news. The RFI is the door.

For a sense of how unusual this is in federal AI funding: the typical DARPA program operates on a three-to-five-year horizon, runs through a single program manager, and is governed by a Broad Agency Announcement that scopes the technical objectives in advance. The typical NSF program operates through merit-reviewed solicitations with letters of intent, full proposals, and program-officer-driven panels. Neither agency, individually, runs a standing forum that pools university capabilities into a single repository and dispatches research projects out of that repository on a rolling basis. AI Forge is doing exactly that — jointly, with NIST's CAISI as a third partner, and with a nonprofit administrator that does not yet exist. (Granted News)

The seventeen-day window between the June 1 announcement and the June 22 RFI deadline is the period in which the forum is building its initial repository of university capabilities. Teams that respond to the RFI are positioning themselves for the first round of forum-dispatched research projects, which the announcement implies will begin in the second half of 2026. Teams that do not respond are not blocked from later forum work, but they are not in the initial repository, and the forum's nonprofit administrator will use the repository as the first source of capabilities when scoping the first projects. The cost of not responding is not a closed door — it is a later entry, on terms the forum has already set.

The three thrust areas, the fifteen challenges, and what the forum is actually funding

AI Forge organizes its research agenda into three thrust areas. The first is AI interpretability — making the behavior, decisions, and impacts of AI systems understandable to humans. The second is AI control — developing tools that can provide strong, verifiable evidence of bounded, auditable, and reliable model behavior. The third is adversarial robustness — building the scientific foundations for AI systems that are not just capable but resilient to deliberate attack.

These three thrust areas are not new. Each has been an active research area in academic AI labs for at least five years, and each has been funded in pieces by DARPA, NSF, the Department of Energy, the Intelligence Advanced Research Projects Activity, and several Department of Defense components. What is new is the consolidation. AI Forge gathers under one forum the research questions that have, until now, been scattered across a half-dozen agency programs with incompatible reporting cycles, divergent intellectual-property terms, and program managers who do not coordinate. The forum's nonprofit administrator will, in theory, set common terms for the research it dispatches, common publication norms, and common access to commercial model and compute resources that no individual academic lab can negotiate on its own.

The fifteen research challenges that the announcement promises to publish — five within each thrust area — are the operational scope. The announcement does not list them in the public press release. The RFI on sam.gov is the document that does, and the RFI is where the specifics of each challenge will be defined for the universities responding by June 22. Teams should treat the RFI not as a procurement notice but as a research-agenda-setting document: the responses inform what challenges the forum prioritizes in its first projects.

The substantive choice of these three thrust areas, rather than (for example) AI capabilities, AI scaling, or AI applications to specific defense problems, is itself a statement. AI Forge is funding the parts of AI research that the commercial sector has the weakest incentive to fund. Frontier labs are racing on capability, scale, and applications. They are not racing — at least not at the same intensity — on interpretability, on formal control guarantees, or on adversarial robustness in the contested environments that matter for national security. The press release is explicit about this. The forum's premise is that many of the AI challenges that matter most for national security "remain underexplored because they lack immediate commercial applications and are not the primary focus of private industry."

The forum model and why it sidesteps normal merit review

The forum model is what makes AI Forge structurally different from either DARPA or NSF's existing AI programs. A traditional DARPA program runs a Broad Agency Announcement, receives proposals, awards contracts to a small number of performers, and runs those performers under a single program manager for the duration of the program. A traditional NSF program runs a solicitation, receives proposals, runs them through merit review, and funds the top-scoring proposals. Both models have built-in delays — months between solicitation and award — and both produce a fixed cohort of performers that is hard to modify mid-program.

The forum model dispatches research projects out of a standing repository of universities that have been vetted in advance. When the forum's nonprofit administrator scopes a new project — say, a six-month sprint on a specific interpretability question — it draws from the repository, assembles a team in weeks rather than months, and runs the project with whatever combination of compute, models, and expertise the project requires. The merit-review function does not disappear; it is absorbed into the up-front vetting of the repository and the project-scoping work done by the forum's governance structure. The forum is, in effect, pre-clearing universities so that future projects can be initiated without each project triggering a full merit-review cycle.

This is a substantial departure from how NSF normally operates. NSF's merit review is the agency's defining institutional commitment, and the agency has historically resisted mechanisms that would let funding flow without per-project review. The fact that NSF is co-signing AI Forge — and not just lending its name but contributing administratively to the forum and accepting the forum's project-dispatch model — is a signal about NSF's posture under the current administration. The agency is willing, at least for AI work classified as national security, to put projects through a vetted-repository model that runs in parallel to its traditional merit-review programs.

For university research teams, this has two implications. First, getting into the AI Forge repository is structurally different from getting an NSF grant — it requires a capabilities response (the RFI), not a research proposal. Teams that submit a capabilities pitch focused on what they can do rather than what they want to do are responding to the document the forum has actually asked for. Second, being in the repository is a multi-year asset, not a one-time funded project. A team that lands in the repository in summer 2026 is positioned for forum-dispatched work through at least FY 2028, and possibly longer if the forum's nonprofit administrator builds the venue into a sustained institution.

What the seventeen-day RFI window actually requires

The RFI is open until June 22, 2026. The submission portal is sam.gov, the federal government's standard procurement and assistance solicitation site, which means teams responding need an active SAM.gov registration with a current Unique Entity Identifier. Universities with active federal awards already have this. Teams without active SAM.gov registration should not attempt to register in the seventeen-day window unless they already have a Dun & Bradstreet number, an active CAGE code, and a research-administration office that can confirm institutional authorities — the registration process can take three weeks even when it is uncomplicated, and SAM.gov registration is the gating step the response cannot bypass.

The substantive content of a strong RFI response will, based on the structure of similar federal capabilities-RFIs in 2024 and 2025, prioritize three things: a clear statement of which of the three thrust areas the responding team can contribute to (and which specific challenges within those thrust areas, once the RFI document is read), a description of the team's access to compute and model resources that complement what the forum will supply (rather than duplicate it), and an honest accounting of the team's track record on projects of the type the forum will dispatch. The forum is building a repository of capabilities, not a portfolio of proposals — the response should read as a brochure of what the team can do, not as a pitch for a specific project.

Responses are not, in the technical sense, binding. A capabilities response does not commit the responding team to any specific project, and it does not commit the forum to fund any specific work. The function of the response is to be in the room — to be one of the universities the forum's nonprofit administrator can draw from when it scopes its first projects in the second half of 2026.

What this means for federal AI funding strategy through FY 2028

AI Forge is one of several signals that federal AI funding is consolidating into a smaller number of larger venues. The NSF AI-Ready America hubs solicited under NSF 26-508 fund one state-or-territory coordination hub per jurisdiction. The NSF X-Labs initiative under the milestone-funding model funds quantum and instrumentation research outside the traditional grant mechanism. The DARPA CLARA high-assurance AI program operates under a Broad Agency Announcement with rolling proposals. The pattern is a movement away from many small, broadly-scoped programs and toward fewer, larger, more architecturally distinct venues — each with its own funding mechanism, its own merit-review approach, and its own multi-year horizon.

For a university research team building an AI funding strategy for FY 2027 and FY 2028, this means the strategy is no longer about tracking individual solicitations from individual agencies. It is about getting into the venues that will, in aggregate, control most of the federal AI research dollars over the next two years. AI Forge is one such venue. The seventeen-day RFI window is the door. The forum is the building. Teams that respond by June 22 are inside.