NSF CICI: Up to $12M for Science Cybersecurity by January 20, and AI Data Provenance Now Leads

Most cybersecurity funding protects networks and systems in general. NSF's Cybersecurity Innovation for Cyberinfrastructure (CICI) program is narrower. It funds security for the machinery of research itself: the instruments, data pipelines, campus research networks, and, increasingly, the datasets that train scientific AI models. Proposals that try to make a general security research project look like science-adjacent work tend to fail on that point.

The next deadline is Wednesday, January 20, 2027, at 5 p.m. submitter's local time, according to the CICI program page. The governing solicitation is still NSF 25-531, posted January 3, 2025. It sets a standing deadline of the third Wednesday in January. NSF anticipates $8 million to $12 million for 12 to 20 awards across four tracks, each lasting up to three years, subject to available funds.

Ten weeks is enough time for a strong proposal only if the science partner is lined up now. CICI's review criteria assume the science partner is part of the team from the start, not added at the end.

Four Tracks, and Where the Awards Are

Compared with the previous solicitation, NSF 23-517, the allocation has shifted. Under NSF 23-517, UCSS was the main track, with 8 to 10 expected awards. NSF 25-531 cut it to 3 to 6 and created IPAAI, the AI data-integrity track, with 4 to 6 awards. That gives IPAAI the highest expected award count of the four tracks. TCR fell slightly, from 2–5 to 2–4. Total funding rose only from $8–11 million to $8–12 million, so IPAAI's awards came mostly out of UCSS.

For applicants this matters in both directions. A UCSS proposal now competes for fewer awards. An IPAAI proposal is in a track NSF built deliberately and expects to fund the most. The total also fits roughly three or four TCR awards at the $1.2 million cap, which shows that NSF regards transition-to-practice work as expensive and selective.

IPAAI: The AI Track Has a Specific Scope

NSF's language for IPAAI is narrow. "The primary focus of the proposal should be directed toward the use of AI in science." That excludes general LLM security, adversarial machine learning on benchmark datasets, and deepfake detection for social media. The track addresses the integrity, provenance, and authenticity of scientific datasets used to train or run AI. Examples include cryptographic provenance for telescope survey data feeding a foundation model, tamper-evident lineage for genomic training sets, and authenticity checks on simulation outputs used as synthetic training data.

That fits a larger federal trend. NSF and DOE are spending heavily on AI-ready scientific data through the Genesis Mission and NSF's dataset programs, including the $83 million Integrated Data Systems and Services effort. CICI's solicitation does not mention Genesis or NAIRR. But a reviewer reading an IPAAI proposal in early 2027 will know how much federal money now depends on trustworthy training data. A proposal that names a real AI-for-science pipeline, with a real data producer as a partner and a real threat model, such as data poisoning, silent corruption, or lost provenance across transformations, is responding to that context.

UCSS, RSSD, and TCR

UCSS ($600K) funds security that scientists will actually use, plus collaborative models between security teams and research groups. NSF 25-531 adds language on Security Operations Centers: NSF "especially encourages SOC proposals focusing on small and under-resourced institutions." With fewer UCSS awards available, a shared or regional SOC model for smaller research institutions, such as a consortium of regional universities or a SOC-as-a-service for EPSCoR campuses, matches the solicitation's stated preference closely.

RSSD ($600K) funds curated security datasets drawn from scientific environments: network traces from research networks, attack data from instrument control systems, and similar sources. The bar for access is higher now. Under NSF 25-531, "all curated datasets are required to be shared publicly and made available through established community platforms," following FAIR principles, and that requirement is enforced through a special award condition. If your data partner cannot agree to public release because of privacy, export control, or a facility policy, RSSD is the wrong track.

TCR ($1.2M) moves existing, proven security research into operational cyberinfrastructure. New language also covers secure dataset caching near computation and integrity tied to the data itself. The key word is existing. One of the review criteria is "Building on existing, recognized capabilities." TCR proposals without a track record of deployed prototypes, pilot users, or earlier NSF-funded results that are ready for transition are poorly positioned.

Rules That End Proposals Before Review

• Two-proposal limit per person. An individual may participate as PI, co-PI, or senior personnel "on no more than two CICI proposals." Extra proposals "will be returned without review. No exceptions will be made." At large research universities, where several security groups may each list the same CISO or research computing director, coordinate this internally before December.
• One proposal per collaboration. Multi-institution teams must submit one proposal with subawards. Separate collaborative proposals are not allowed.
• No voluntary committed cost sharing. Don't include it, even as a gesture of institutional support.
• Eligible institutions: U.S.-accredited two- and four-year institutions of higher education with a U.S. campus, and U.S. nonprofit, non-academic organizations "directly associated with educational or research activities." That covers independent research institutes and facility operators but not commercial security vendors as prime applicants.

There is no limit on proposals per organization, which means coordination among PIs is the only constraint.

The Review Criteria Reward Science Partners

In addition to Intellectual Merit and Broader Impacts, CICI applies seven solicitation-specific criteria: science-driven, innovation, open and FAIR datasets, close collaborations among stakeholders, building on existing, recognized capabilities, project plans and architecture, and sustained impact.

Four of the seven depend on the relationship with scientists. The solicitation also strengthened program-wide requirements. NSF 23-517 said proposals "should include" items such as a sustainability plan. NSF 25-531 says they "must include" them, and adds a required discussion of "any ethical and operational concerns of the work." A new Letters of Collaboration section gives a clear place to document science partners.

In practice, the science team should write part of the problem statement. A letter that says only "we are excited to collaborate" will not satisfy the "close collaborations" criterion. A letter from a facility director that describes the specific pipeline, the incident or near-miss that prompted the work, and the staff time the facility will contribute is much stronger. The sustainability plan should name who operates the tool after year three, such as campus research computing, a facility's operations budget, or a community software foundation, and whether that operator has agreed.

A Ten-Week Calendar

• Now through Thanksgiving: Choose the track. Use the table above to estimate your odds honestly. Confirm the science partner and the data release terms, especially for RSSD and IPAAI.
• Early December: Check every senior person against the two-proposal limit across your institution. Email the program officers, Daniel Massey and Kevin Thompson, with a one-paragraph summary if you're unsure which track fits.
• Mid-December: Draft the threat model and architecture, which the "project plans and architecture" criterion evaluates directly. Request letters of collaboration before the holidays, since facility directors are often hard to reach over the break.
• Early January: Finish the sustainability plan, the ethics and operational-concerns section, and the data management plan with FAIR and public-release commitments.
• January 13: Submit through Research.gov with a week to spare.

CICI is a small program, about $10 million a year across all of science cyberinfrastructure, but each award comes with credibility among research computing staff and facility operators. If you're turning a security prototype and a science partnership into a three-year plan, Granted can help you organize the narrative around the seven review criteria before January 20.