DARPA's Widest Open Door Closes November 1. The I2O Office-Wide BAA Abstract Window Has Two Weeks Left.

Almost every DARPA opportunity you can name is a program. A program manager has a thesis, writes a solicitation around it, and funds teams that advance that specific thesis. If your idea does not fit one of the roughly two dozen named programs running at any moment, there is nothing to apply to.

Office-wide Broad Agency Announcements are the exception, and they are structurally different in a way that matters enormously for anyone with an unsolicited idea.

DARPA's Information Innovation Office — the home of the agency's artificial intelligence, cybersecurity, autonomy, and information science portfolio — runs its office-wide BAA under solicitation number HR001126S0001, posted November 28, 2025. It exists specifically to catch research concepts that fall outside the scope of every active program. The BAA's own framing is that it seeks ideas "not currently covered by ongoing I2O programs."

That door closes to new abstracts on November 1, 2026 at 5:00 p.m. Eastern. Invited full proposals are due November 30, 2026 at 5:00 p.m. Eastern.

As of today, that is roughly two weeks of abstract window remaining on a solicitation that has already lost several months of its intended life.

The pause is the story

An office-wide BAA is normally a twelve-month instrument. Posted in late November 2025, HR001126S0001 should have accepted abstracts continuously through the fiscal year on a rolling basis — submit whenever your idea is ready, get assessed as it arrives.

That is not what happened. The BAA was temporarily paused effective May 21, 2026, tied to organizational changes that included an update to the office's name and a broader realignment of strategic priorities. No new abstracts were accepted during the pause, though abstracts already submitted remained under review. An updated BAA was expected to resume accepting abstracts in August 2026.

Work out the arithmetic on the usable window:

For anyone who developed an idea between late May and August — which is most of the summer, including the window when many university teams actually have time to write — the only live submission opportunity is the one closing now.

There is a secondary consequence that is easy to miss. The queue of abstracts that accumulated before the pause did not disappear; it sat under review while the office reorganized. New abstracts arriving in October are landing in a review pipeline that is simultaneously working through a backlog and absorbing a refreshed set of strategic priorities. That is not a reason to skip submitting. It is a reason to be unusually explicit about which current priority your idea serves, because the people reading it have recently been told their priorities changed.

One practical caution before you act on any date in this post. Because of the rename and realignment, the office-wide solicitation now appears in federal listings under more than one office label, and at least one commercial aggregator carries a June 2026 response date that reflects the pause rather than the current window. Pull the active version on SAM.gov under solicitation number HR001126S0001 and confirm the deadline against the posted document before you build a schedule around it. Office-wide BAAs get amended, and the amendment is the authority — not the summary.

What the four thrust areas actually signal

The BAA organizes its interests around four thrust areas. Read them as a map of what the office will and will not find relevant, because relevance is the first screen your abstract has to pass.

Transformative AI. Trustworthy, disruptive AI technologies and methodologies relevant to national security, with emphasis on explainability and ethical alignment. The operative word is disruptive. This is not the thrust for a better benchmark score.

Resilient, adaptable, and secure software and complex systems. Formal methods and third-wave AI applied to understanding, building, updating, repairing, and restoring complex software and cyber-physical systems. Formal methods are named explicitly, which is a meaningful signal — this thrust has a genuine appetite for proof-carrying approaches rather than empirical hardening alone.

Offensive and defensive cybersecurity and privacy. AI-leveraged trustworthy cyber capabilities, spanning both directions.

Fighting in the information domain. Measuring the health of and detecting attacks on the information domain, across cognitive, semantic, tracking, and transport levels. This is the thrust that has the least overlap with civilian research funding and consequently the thinnest competition from academic teams who have never thought about it.

What DARPA explicitly excludes is at least as useful as what it invites: research yielding evolutionary improvements to existing state of the art, and work already funded or completed elsewhere. The phrase "incremental or evolutionary improvements will not be considered" is not boilerplate hedging. It is the rejection criterion that disposes of most abstracts.

The abstract is the competition

The structural fact that governs everything about applying here: this is a mandatory two-step process, and a proposal will not be reviewed unless an abstract has been submitted first.

You submit an abstract. The office reviews it for relevance and interest. Only proposers who receive an official invitation may submit a full proposal. There is no path where a strong full proposal rescues a weak abstract, and no path where you skip the abstract because you are confident in your technical volume.

This inverts the usual effort allocation. On a standard federal solicitation, the proposal is where 90% of the writing labor goes. Here, the abstract is the gate that eliminates people, and it is the document that is cheap to write badly.

Three things follow from that.

First, write the abstract to be rejected on the right grounds. The reviewer's question is not "is this good research?" It is "is this relevant to a current thrust, and is it revolutionary rather than evolutionary?" An abstract that buries its thesis in background, or that describes a capable team and a sound method without stating plainly what becomes possible that is impossible today, fails on relevance even when the science is strong.

Second, submit early inside the rolling window, not on the deadline. Rolling assessment means abstracts are read as they arrive. An abstract submitted in mid-October gets read against a less congested queue than one submitted on November 1, and — this is the part people miss — an early submission that draws an invitation gives you materially more than 29 days to write the full proposal.

Third, respect the 29-day compression. November 1 to November 30 is the maximum interval between invitation and full proposal, and most invitations will arrive with less runway than that. A full DARPA proposal is a technical volume, a cost volume, and a summary slide. If your plan is to begin the cost volume after the invitation arrives, you are planning to assemble a DARPA cost proposal in under four weeks with a Thanksgiving holiday inside it. Build the budget now, against the abstract you are submitting, and hold it.

Award size, instrument flexibility, and who can apply

DARPA does not publish fixed award amounts for office-wide BAAs, and it anticipates funding a limited number of proposals, with budgets determined by technical approach, scope of work, and alignment with office priorities. Third-party estimates for I2O office-wide awards cluster in the $500,000 to $5 million range, with some analyses extending to $10 million. Treat those as orientation, not as a ceiling to budget against.

Because the amount is unposted, cost realism carries unusual evaluation weight. There is no ceiling to anchor to, which means your budget is judged against your own technical scope rather than against a number someone else chose. An inflated budget on an office-wide BAA is a self-inflicted wound in a way it is not on a solicitation with a published cap.

The instrument flexibility is the underappreciated asset here. This BAA can award through procurement contracts, grants, cooperative agreements, and Other Transaction agreements for prototypes or research. DARPA invites red-lined edits to model contracts and OT agreements. For a small company or a nontraditional performer, that means the terms are negotiable in a way that a standard grant program's terms are not — and that the right instrument for a prototype effort may not be a grant at all.

Eligibility is broad: U.S. businesses, universities, nonprofits, and federally funded research and development centers, with restrictions applying to foreign entities and FFRDC participation. The administrative codes are NAICS 541715 and PSC AC11, and the BAA coordinator is reachable at hr001126s0001@darpa.mil.

The decision to make this week

If you have a concept in AI trustworthiness, formal-methods software assurance, AI-enabled cyber, or information-domain measurement that no named DARPA program covers, the question is not whether your idea is ready. It is whether a two-to-four page abstract stating the idea, the discontinuity, and the thrust it serves can be written in the next ten days.

If yes, write it and submit it before November 1. The downside is a short document and a no.

If no, the office-wide BAA is the one DARPA instrument that reliably reissues. Office-wide solicitations are annual by construction, which means a new number — and a fresh twelve-month rolling window — typically appears at the start of the next cycle. That is the fallback, and it is a real one. But it is also a full year of waiting for an idea that could have cost you four pages this month, and the FY2026 cycle has already demonstrated that a posted twelve-month window is not a guaranteed twelve months of access.

Verify the active solicitation on SAM.gov, confirm the November 1 abstract date against the current amendment, and write the abstract first.