ARPA-H's SURPASS Wants to Cut Drug Trials From a Decade to Four Years. The Solicitation Asks for Something Almost No Single Organization Has.

The Advanced Research Projects Agency for Health announced the SURPASS program on September 30, and the press coverage focused on the headline: shorten clinical development from more than a decade to under four years, using artificial intelligence and simulation. That framing is accurate and almost useless for anyone deciding whether to compete. The actual story of SURPASS is in the Innovative Solutions Opening, ARPA-H-SOL-26-164, where the agency has written an eligibility architecture that filters out nearly every organization that would normally respond to an ARPA-H clinical trials solicitation.

SURPASS — Simulation-augmented, Real-time Platform Adaptive Seamless Trials — is run out of ARPA-H's Proactive Health Office under program manager Daria Fedyukina. The Solution Summary is due November 30, 2026 at 5:00 p.m. ET, and submitting one is mandatory: there is no path to a pitch package without it. An informational webinar runs October 15 with registration closing October 12. The Proposers' Day sits in the week of November 6 in Washington, with a virtual option and an October 28 registration cutoff. Questions close November 2. Invited teams submit a Solution Pitch Package estimated for January 22, 2027, with virtual pitches in early February.

That is roughly eight weeks from announcement to the gating submission, over a stretch that includes Thanksgiving. For a program that requires a consortium nobody has assembled yet, eight weeks is the first real constraint.

What SURPASS Is Actually Asking For

The program is organized into three technical areas, and critically, they are not separable tracks that different teams can bid independently. A responsive proposal addresses all three.

Technical Area 1 — the phaseless design engine. SURPASS wants a simulation environment that runs adaptive platform trials without conventional Phase 1 / Phase 2 / Phase 3 separation. The inputs are digital twins, synthetic control arms, and disease progression models. The premise is that if you can simulate patient trajectories well enough, you can collapse the dose-finding, efficacy-signal, and confirmatory stages into one continuous enrollment with prespecified adaptations rather than three sequential protocols with two multi-year restart penalties between them.

Technical Area 2 — continuous inference. The statistical problem created by TA1 is that conventional frequentist inference is not valid when you look at the data continuously and change the trial while it runs. TA2 asks for an inference engine that stays valid under arm additions, arm removals, and seamless stage transitions — explicitly naming Bayesian, group-sequential, and anytime-valid approaches. This is the technical core of the program and the piece most likely to determine which teams survive review.

Technical Area 3 — the agentic operations layer. AI systems that handle Master Protocol documentation, site activation, regulatory workflow, enrollment, and the mechanical implementation of adaptations. The solicitation is unusually direct that this does not relocate accountability: AI-enabled automation does not transfer or replace the responsibilities of sponsors, investigators, clinical sites, statistical committees, data safety monitoring boards, institutional review boards, or regulatory authorities. That sentence is the agency pre-empting the obvious objection, and it is also a design requirement — the operations layer has to be auditable by every one of those bodies.

The Structure That Decides Who Can Compete

Here is where SURPASS departs from the ARPA-H template. The program runs 60 months in two stages with a hard gate between them.

Stage I is 24 months. The performer has to build the integrated capability, establish a Master Protocol, open its first clinical site by Year 2, and onboard three intervention arms. Stage II is 36 months of prospective validation — actually running adaptations in a live trial and reaching a minimum of six clinical sites.

The continuation gate is the part worth reading twice. To proceed from Stage I to Stage II, the team must obtain FDA authorization to proceed for both the platform Master Protocol IND and at least one investigational arm. That is not a technical milestone ARPA-H can grade leniently. It is a third-party regulatory decision, inside 24 months, on a trial design the agency is simultaneously asking the team to invent. Teams that have never held an IND as a sponsor should price that risk honestly before committing.

The composition requirements are equally specific:

• At least two for-profit drug or biologic sponsors on the team, and at least one of those must be a startup or small company.
• At least three intervention arms proposed, of which at least two must be interventions without a prior approved indication, and at least one must have no human data at all.
• A minimum of six clinical sites by program end, with the first open by Year 2, and one site holding the Master Protocol IND.
• A lead platform operator with the expertise, infrastructure, authority, and long-term incentive to sustain a multi-sponsor platform after federal funding ends.
• FFRDCs are generally prohibited as primes or subperformers; government entities are generally ineligible with narrow exceptions. Active SAM registration and a Unique Entity Identifier are prerequisites.

Disease area is open, with the condition that it demonstrate substantial unmet need and realistic platform sustainability. Neurodegenerative disease is called out as of particular interest.

Read those constraints together and the profile of a winning team emerges: an academic medical center or disease-focused network with platform trial experience, acting as lead operator; two pharmaceutical sponsors willing to put unapproved assets into a shared protocol they do not individually control; a biostatistics group that can defend anytime-valid inference to FDA; and an AI engineering capability for the operations layer. The hard part is not any single component. It is that the two sponsors have to agree to pool a control arm and cede adaptation decisions to a shared committee — a commercial concession, not a scientific one, and the reason most platform trials in oncology and Alzheimer's took years to stand up.

The Funding Shape Nobody Publicized

ARPA-H did not disclose a program budget, and the trade press noted that absence as a transparency problem. The solicitation itself is more informative than the press release. ARPA-H anticipates fully funding the trial expenses associated with the first three intervention arms, may provide partial funding for the fourth and fifth, and does not anticipate funding beyond the fifth arm.

That is a sustainability test disguised as a budget line. The agency is paying for the platform to prove it works on three arms and signaling that arms six onward must be sponsor-funded. A proposal that assumes federal money for arm eight is non-responsive. A proposal that specifies the per-arm cost-recovery model the platform will charge sponsors in Year 4 is answering a question the solicitation asks obliquely but grades on.

Awards are Other Transaction agreements, multiple awards anticipated, selected on what ARPA-H determines most advantageous to the government considering innovation, feasibility, and scientific and technical merit. The OT mechanism matters practically: no standard F&A negotiation, milestone-based payments, and terms negotiated individually. Organizations whose sponsored programs office has only ever processed grants should start the internal conversation now, not in January.

The Submission Ladder

The Solution Summary is four narrative pages. In those four pages ARPA-H expects the disease area, the intervention arms, the trial design, the innovations across all three technical areas, the regulatory strategy, the sustainability model, and the competitive differentiation. That is an extraordinary compression ratio, and it means the summary is not a teaser — it is the proposal, abbreviated. Teams that treat it as a letter of intent will not advance.

Invited teams receive a feedback letter confirming the Pitch Package due date. The Pitch Package is a 21-slide maximum deck, an Excel Gantt chart, a preliminary Master Protocol with adaptive design details, a two-page description of the in silico and AI models, a Scope of Work and Cost Workbook, and an optional regulatory appendix. The preliminary Master Protocol is the tell: ARPA-H is asking for a draft of the document that will eventually go to FDA, at pitch stage, which is why the regulatory strategy section of the four-page summary carries more weight than its length suggests.

Strategy

If you are a potential lead operator: your single highest-value action in October is securing the two for-profit sponsors with letters that commit named assets, not categories of assets. The solicitation requires two interventions without an approved indication and one with no human data. A sponsor letter that says "we are interested in contributing a candidate" fails that test. One that names the molecule and its current stage passes it.

If you are a startup with a preclinical asset: you are the scarcest input in this competition, not the most vulnerable one. The requirement that at least one for-profit sponsor be a startup or small company, combined with the requirement that at least one intervention have no human data, describes you specifically. Reach out to platform trial operators in your disease area rather than waiting to be recruited.

If you are a biostatistics group: TA2 is where proposals will be won and lost, and anytime-valid inference under arm addition is a narrow specialty. Being named on more than one team is worth evaluating against exclusivity pressure.

If your expertise is only in the AI operations layer: TA3 alone is not a proposal. The realistic path is subcontracting into a consortium, and the October 15 webinar plus the November Proposers' Day are the two structured opportunities to find one.

SURPASS fits into a cluster of ARPA-H and HHS efforts — STACK, COMMONS, and CINCH on infrastructure and data — and alongside NIH's parallel push to replace animal models with human-based methods under Complement-ARIE. The common thread is a federal bet that the preclinical-to-clinical pipeline fails for structural reasons, not for lack of effort. SURPASS is the clinical-trial-side wager. Its eligibility rules make clear that ARPA-H knows the obstacle is organizational, not computational: the simulation engine is buildable, and getting two drug companies to share a control arm is the research problem.

Teams that understand that distinction have until November 30.