NASA Promised ROSES-2026 in July. It Is August 29, ROSES-25's Rolling Deadlines Hit the Wall Monday, and the New Solicitation Still Reads 'Pending.'

In January 2026, NASA's Science Mission Directorate told the research community something reassuring. ROSES-2025 had been released in July 2025 instead of the traditional February, and rather than snap back, SMD would make July permanent: "For simplicity, consistency, and to maintain the annual schedule, SMD plans to release ROSES-2026 in July again this year rather than February." ROSES-2025 would "continue to solicit until ROSES-26 is released," and — critically — "ROSES-25 will be amended soon to extend the close date for the 'no due date' and 'flexible' due date program elements."

July came and went. As of today, ROSES-2026 (NNH26ZDA001N) exists as a record in NSPIRES with a status of "Pending." No release notice, no Summary of Solicitation, no Table 2 or Table 3 of due dates.

Meanwhile ROSES-2025 keeps going. Its NSPIRES record now carries a close date of December 31, 2026 and an amendment log running past Amendment 69 (D.9, Habitable Worlds Observatory Instrument Concept Assessments, issued in the last week of August). SMD has continued to issue corrections and clarifications through August 27.

This is the fourteenth month of a solicitation designed to run twelve, and the practical consequences land on Monday.

The August 31 cliff is real for three program elements

Amendment 39, issued January 26, 2026, was the extension SMD promised. It moved three rolling or flexible program elements to new outer limits:

• C.2 Solar System Science — extended to August 1, 2026 (already passed)
• A.4 Rapid Response and Novel Research in Earth Science — extended to August 31, 2026
• A.10 INNOVATE — extended to August 31, 2026

A.4 and A.10 are the two elements most likely to be sitting half-drafted on someone's desk right now, precisely because they were the ones with no fixed due date. A.4 exists to fund science that cannot wait for an annual cycle — a wildfire, a volcanic eruption, an unexpected satellite observation. A.10 INNOVATE is the Earth Science technology and novel-methods lane. Both were built for opportunistic submission, and both have been the release valve for anyone who missed a fixed date elsewhere in ROSES-25.

On September 1, that valve closes, and there is no posted successor. ROSES-2026 has not been released. Until it is, a researcher with a time-critical Earth Science idea has no rolling NASA mechanism to submit into.

If you have anything close to submittable in A.4 or A.10, submit it this weekend. A thin proposal into an open element beats a polished one into a closed one, and the ROSES-25 close date on the parent solicitation — December 31 — does not extend individual program elements past their own stated dates. That distinction has tripped up proposers before: the omnibus stays open, your appendix does not.

The generative-AI rule that arrived mid-solicitation

On July 30, 2026, SMD amended A.14 Atmosphere with several corrections, one of which added a new Section 4.3.3, "Use of Generative AI." It is a small insertion with real teeth, and because it arrived fourteen months into the solicitation, a large fraction of the community drafted proposals against a version of the document that did not contain it.

NASA's position, grounded in its Scientific and Research Integrity Policy (NPD 1920.1A), is permissive but conditional. SMD does not prohibit generative AI tools in proposal preparation or award reporting. What it requires is citation. When a proposal uses generative AI, the citation must include:

1. The name of the program used
2. The version number of the generative AI used
3. The date and time the tool was used to generate proposal text
4. A statement of how the tool was used

Three additional constraints matter more than they appear:

Human professional writers must also be acknowledged. The disclosure obligation is not AI-specific; it extends to any non-team writing assistance. Institutions that route proposals through a central editing office should check whether that office has ever been named in a NASA submission.

Unacknowledged AI-generated content is discouraged — the operative risk being that undisclosed use, if discovered, becomes a research-integrity matter rather than a proposal-formatting matter. The proposal team "bears full responsibility for accuracy and authenticity" of everything submitted, which forecloses the defense that a hallucinated citation came from the tool.

ITAR- and EAR-controlled information cannot be entered into publicly accessible AI systems. This is the clause with genuine institutional exposure. A researcher pasting an instrument specification into a consumer chatbot to tighten the prose may have committed an export-control violation independent of anything in the proposal. Any lab working on flight hardware should treat this as a compliance rule, not a style guideline.

Note also that ROSES uses dual-anonymous peer review (DAPR) as the default. DAPR and AI disclosure interact awkwardly: an AI-use citation that names an institutional tool or a lab-specific model can leak identity into an anonymized document. Put the disclosure where the DAPR guidelines put identifying material — not in the anonymized science section.

The $0.09 charge that has to be in your budget

The quieter change is a line item. ROSES-25's A.1 Earth Science Research Overview was updated to require that Earth Science proposals planning to use NASA high-end computing resources include those costs in the proposal budget, at a rate of $0.09 per SBU (Standard Billing Unit).

For a decade, NASA HEC allocations functioned as an in-kind resource: you requested time, you received an allocation, and it did not appear in your budget. Making it a priced line item does three things at once.

It shrinks the science you can buy with a fixed award ceiling. A computationally heavy Earth Science project that previously spent its entire budget on people now spends part of it on compute, at a rate set outside the proposal.

It forces an early architecture decision. Proposers must now estimate SBU consumption at proposal time — before the code is tuned, before the resolution is final. Underestimate and you exhaust the line mid-project; overestimate and you have surrendered budget you could have spent on a postdoc.

It creates a comparison NASA has not previously invited. Once HEC time carries a price, a proposer can rationally compare it against commercial cloud. For some workloads the answer will be that commercial compute is cheaper, and the proposal that says so with numbers will read as more rigorous than the one that defaults to NASA resources out of habit.

Anyone whose ROSES-25 Earth Science proposal was drafted before this update and does not carry a compute line should add one now. A budget that omits a required cost category is a compliance problem, not a competitiveness problem.

What to do in the next 72 hours, and then in September

This weekend. If you have an A.4 Rapid Response or A.10 INNOVATE proposal in any submittable state, file it before August 31. Confirm the exact close time in the program element PDF — ROSES dates are typically 11:59:59 PM Eastern, but the amended dates are the ones that govern, and NSPIRES enforces them literally.

Before you submit anything into ROSES-25. Pull the current program element PDF from NSPIRES rather than the copy in your shared drive. With 69-plus amendments, the odds that your local copy is current are poor. Check three things specifically: whether Section 4.3.3 has been added to your element, whether a computing-cost requirement applies, and whether the due date in your calendar matches the amended date.

In September. Watch NSPIRES and the SARA notices for the ROSES-2026 release. When it lands, the two documents to read first are the Summary of Solicitation — where the cross-cutting policy changes live — and Table 3, which lists every program element with its due dates. Whether the generative-AI section and the SBU charge appear as ROSES-wide policy rather than element-by-element patches is the single most consequential thing to check.

Plan for a compressed cycle. If ROSES-2026 releases in September or October rather than July, every downstream date compresses: Step-1 due dates, Step-2 due dates, panel scheduling, and award start dates. Programs that historically gave you four months between release and Step-1 may give you six weeks. The teams that fare well in a compressed cycle are the ones that drafted against last year's element text and adapted, not the ones that waited for the new PDF to start thinking.

NASA's science budget environment makes the timing more than an inconvenience — the FY2027 request proposed deep cuts across NASA, NOAA, and USDA research, and a solicitation that slips is a solicitation whose budget lines are still being argued over. The community's leverage in that argument is a full proposal pipeline. Which starts with getting Monday's submissions in.

For general ROSES questions, SMD directs inquiries to Max Bernstein at SARA@nasa.gov; program-element questions go to the point of contact named in each element's PDF.