Every DARPA AI Program You Can Apply To in 2026

DARPA does not publish a tidy list of open AI programs. It publishes a page of office-wide BAAs refreshed annually, a rolling feed of program-specific solicitations on SAM.gov, and a research programs index that mixes active programs with ones that wrapped up years ago. If you are a university lab, independent research organization, or startup trying to figure out which DARPA AI opportunities are actually open right now, the answer requires digging through multiple systems at once.

Here is what is actually open as of early 2026, what closed, and what the path to a proposal looks like.

The I2O Office-Wide BAA: DARPA's Broadest AI Door

The Information Innovation Office (I2O) is DARPA's home for AI, cybersecurity, autonomy, and information science. Its FY2026 office-wide BAA — solicitation number HR001126S0001, released November 28, 2025 — is the widest entry point for AI researchers who have a breakthrough idea but no specific named program to attach it to.

The I2O BAA organizes its interests around four thrust areas: transformative AI (trustworthy, explainable, ethically-aligned systems); resilient and secure software; offensive and defensive cybersecurity; and what DARPA calls fighting in the information domain — cognitive and semantic operations, digital artifact tracking, adversarial influence. All four have AI threads running through them.

Abstract deadline: November 1, 2026 at 5:00 PM ET. Full proposals are due November 30, 2026. Abstracts are required before full proposals — I2O reviews them for relevance and only invited abstracts proceed to full proposal stage. There is no fixed award floor or ceiling stated in the BAA, but individual awards on office-wide BAAs typically range from $500,000 to $5 million depending on scope. Proposals must originate as novel research concepts, not incremental extensions of existing projects.

The parallel office-wide BAA from the Defense Sciences Office (DSO), solicitation HR001124S0039, also runs on a rolling basis and is worth checking for AI-adjacent research in physical and computational sciences.

CLARA: The High-Assurance AI Program Still Accepting Proposals

The most concrete named AI program with an open solicitation right now is CLARA — Compositional Learning-And-Reasoning for AI Complex Systems Engineering. Published under solicitation number DARPA-PA-25-07-02, CLARA is a fundamental research program that aims to integrate algorithmic reasoning (AR) with machine learning (ML) into systems that are demonstrably trustworthy, not just empirically good.

The gap CLARA targets is well-documented: modern ML systems achieve impressive benchmark performance but cannot explain their reasoning, fail unpredictably in edge cases, and resist formal verification. CLARA wants approaches that combine the logical rigor of formal reasoning methods with the pattern-recognition capability of ML — tight integration, not loose coupling. The program is structured in two technical areas: TA1 funds development of new high-assurance ML/AR composition approaches (theory, algorithms, open-source code), while TA2 builds a software composition library that integrates validated TA1 tools into a common framework.

Awards are up to $2 million over 24 months. Full proposals are due April 10, 2026. DARPA held an information session via webcast on February 19, 2026. The agency's stated goal is to execute awards by June 9, 2026 — within 120 days of posting. Eligible proposers include universities, research organizations, and small businesses. All software deliverables must be released under a permissive open-source license.

CLARA lives at DARPA's Defense Sciences Office, and it is the clearest on-ramp for researchers working at the boundary of formal methods and machine learning.

ML2P: Making AI Energy-Aware

Less visible but technically significant is the Mapping Machine Learning to Physics program, or ML2P, running under solicitation DARPA-PS-25-32 through I2O. The program's premise is that energy consumption is not a first-class citizen in how ML models are currently trained, evaluated, or deployed. ML2P wants to change that by mapping model performance to physical electrical characteristics measured in joules — creating models that understand and can optimize their own power draw.

The total ML2P budget is anticipated at or below $5.9 million across multiple performers ($3.5 million in Phase 1, $2.4 million in Phase 2). Eligibility is broad: large and small businesses, research institutions, universities, and non-traditional defense contractors. FFRDCs and UARCs may apply only with prior approval. Like CLARA, all tools produced must be released under a permissive open-source license.

Abstracts were due October 2025 and proposals closed December 8, 2025. ML2P is now in evaluation. It is included here because researchers who missed this cycle should watch for follow-on work — the open-source requirement means the tooling will be publicly available, and successor programs are likely.

SABER: Red-Teaming Military AI (Classified Program)

The Securing Artificial Intelligence for Battlefield Effective Robustness program — SABER, solicitation HR001125S0009 — closed its proposal window in summer 2025 (the original May 6 deadline was extended to June 3, 2025). SABER is a Secret collateral program and is notable because BAE Systems has already received a SABER contract, suggesting awards are underway.

SABER's goal is to build an operational AI red team equipped to assess battlefield AI systems against data poisoning, adversarial patches, model stealing, and electronic warfare attacks. The structure involves 24-month cycles of AI security operational test and evaluation exercises (called SABER-OpX), with contractor teams providing attack techniques, tools, and red teaming frameworks.

SABER is closed to new proposals for this cycle. But the program signals DARPA's institutional direction: the agency is building permanent, repeatable infrastructure for adversarial AI assessment — not just publishing research — and expects that infrastructure to expand. Organizations with capabilities in AI security, red teaming, or electronic warfare should monitor I2O's solicitation feed for follow-on work.

The GARD Chapter Is Closed

Guaranteeing AI Robustness Against Deception — GARD — was DARPA's earlier adversarial AI program and frequently appears in searches alongside newer I2O programs. Its four years of work produced the Armory evaluation platform, the Adversarial Robustness Toolbox (ART), and APRICOT datasets, all released as open source and still widely used in the research community. The GARD program page remains active on DARPA's website, but the program concluded in early 2024. No new proposals are being accepted.

GARD's tools are worth knowing. The Armory testbed — available on GitHub — provides repeatable, scalable evaluation of adversarial defenses. ART provides a comprehensive toolkit for defending and evaluating ML models against adversarial attacks. Both are relevant infrastructure for any organization building an AI robustness program, regardless of whether you are pursuing DARPA funding.

AI Forward: The Successor to AI Next

DARPA's AI Next campaign — announced in 2018 with more than $2 billion in investment across roughly 50 programs — formally gave way to the AI Forward initiative, which received $310 million in the FY2025 budget request. AI Forward's organizing principle is trustworthiness: the practical question of how to build AI systems that reliably operate in adversarial, ambiguous, and high-stakes environments.

AI Forward is not itself a solicitation; it is an umbrella under which DARPA's I2O office organizes a series of AI Exploration opportunities (AIEs). AIEs are deliberately fast — DARPA uses streamlined contracting to target a start date within three months of announcement. They appear in I2O's solicitation feed with relatively short response windows, sometimes 30 to 45 days from announcement to abstract deadline.

The practical implication: if you are tracking I2O for AI programs, the AIE mechanism is where the fastest-moving opportunities will show up. Subscribe to I2O updates on DARPA's website and set a SAM.gov watch on DARPA's I2O as the contracting office.

How BAA Responses Actually Work

DARPA BAA proposals are not grant applications in the NIH or NSF sense. A few distinctions matter.

Abstracts come first. Nearly every I2O program — office-wide and named — requires an abstract (typically 3-5 pages) before a full proposal. I2O reviews abstracts and issues explicit invitations to submit full proposals. Submitting a full proposal without an invitation is possible but signals a misunderstanding of how DARPA operates.

Contact program managers before submission. DARPA Proposers' Days (published on DARPA's events feed) give direct access to PMs. They are expected to hear from potential performers. A PM who tells you your white paper is not a fit for the current program is saving you significant effort. One who expresses interest before abstract submission is a strong signal worth acting on.

Technical metrics over impact statements. DARPA proposals are evaluated on technical merit, innovation relative to state of the art, and the team's ability to execute. The structure is: what are you going to do, what is novel about your approach, what are the specific milestones, and what are the measurable outcomes? Significance sections that work at NIH do not translate.

Transition matters. Every proposal needs to address how research outputs transition to operational capability. This does not mean promising a fielded system — it means demonstrating that you have thought about the acquisition pathway and the eventual military end-user.

What to Watch Through 2026

Three things to track over the coming months: the I2O office-wide BAA (HR001126S0001) runs through November 2026 and is the widest open door for AI research. The Young Faculty Award 2026 solicitation — DARPARA2502, posted to SAM.gov and Grants.gov — targets early-career researchers and explicitly includes interpretable reinforcement learning, logical AI, and AI knowledge representation among its relevant topic areas. And AIE announcements from I2O, which can appear at any point and close quickly.

DARPA has never been a volume funder. Program managers select a small number of performers they believe can execute high-risk research. The organizations that win are those that build relationships with the relevant office before BAA release, submit abstracts that demonstrate genuine technical novelty, and know the transition landscape well enough to articulate where their work lands.

Granted's defense research funding search pulls open DARPA solicitations alongside DOD SBIR, ONR, AFOSR, and ARO opportunities — so you can track what is actually open without monitoring five separate portals.