United States - National AI Research Pilot

Overview

The National AI Research Resource (NAIRR) pilot is a U.S. National Science Foundation (NSF) led proof-of-concept, launched in January 2024, that provides a federated platform of compute, datasets, models, software, training, and user support to the U.S. research and education community. The pilot implements the phased roadmap of the NAIRR Task Force Final Report (Jan. 2023) and seeks operational lessons to inform a sustainable national capability. The pilot aggregates in-kind contributions from federal agencies and non-governmental partners and operates a coordinated allocation process, a user portal, focused tracks for secure and privacy-preserving research, and education/classroom programs. See the NSF NAIRR pilot page for program descriptions and partner lists: NAIRR pilot (NSF).

Definitions

This policy entry adopts operational definitions used by the NAIRR pilot and related documents. "NAIRR Pilot" denotes the January 2024 pilot-phase infrastructure and activities administered by NSF and its PMO. "NAIRR-OC" refers to the NAIRR Operations Center to be established via solicitation NSF 25-546. "Resource Providers" are federal agencies, academic centers, cloud and model vendors, and nonprofits contributing compute, datasets, models, or tools. "Allocations" are time-, compute-, data- or service-limited awards made to researchers via NAIRR processes. "NAIRR Secure" is the pilot focus area for privacy-preserving and controlled-access research. "PETs" means privacy-enhancing technologies. These terms align with the NAIRR Task Force report and NSF solicitations and program guidance.

Governance and Institutional Framework

Governance during the pilot is multi-layered. NSF hosts a Program Management Office (PMO) that administers award mechanisms and operational coordination, while a federal Steering Committee provides strategic direction and interagency coordination. The NAIRR pilot invites contributing partners into a cooperative stewardship model where operational responsibilities are divided among NSF, federal agency resource owners, the NAIRR-OC (once awarded), and community partners. Oversight mechanisms include award terms and cooperative agreements; technical guidance and standards are supplied by agencies such as NIST and by working groups focused on security, data stewardship, fairness and civil rights. Community input is solicited through public workshops, RFI processes, and pilot user engagement forums. Relevant program pages and the NAIRR-OC solicitation describe roles and responsibilities in detail: NSF solicitation NSF 25-546 and the pilot overview at NAIRR Pilot portal.

Key Focus Areas

The NAIRR pilot concentrates on four operational focus areas: (1) NAIRR for Research — enabling access to large-scale compute and models to support fundamental and translational AI research; (2) NAIRR Secure — co-led by NIH and DOE to assemble privacy-preserving resources and controlled-access biomedical datasets; (3) NAIRR Software and Data — enabling interoperable software, datasets, and evaluation tools; and (4) NAIRR Classroom — expanding inclusive AI education and training. Across these tracks, the pilot prioritizes: equitable access (including outreach to Historically Black Colleges and Universities and Minority-Serving Institutions), privacy and civil-rights safeguards (data-use agreements, PETs), testing & evaluation (benchmarks, red-teaming and performance metrics), and technical interoperability (common portal, APIs and metadata standards). The pilot uses partner-contributed offerings (cloud credits, model access, HPC allocations) and funds demonstration projects to explore best practices for resource allocation, reproducibility, and responsible research. The pilot's operational choices are designed to surface governance, security, and sustainability lessons for a long-term NAIRR while ensuring the day-to-day needs of diverse research projects — from climate modeling to biomedical AI and education — are served.

Implementation Framework

Implementation of the pilot is built on a mix of NSF awards (research and infrastructure), Dear Colleague Letters soliciting datasets and projects, partner-contributed in-kind resources, and competitive solicitations for the NAIRR-OC. Core elements include a centralized portal that catalogs available resources and manages allocation requests; allocation committees and review processes that vet requests against eligibility and safety criteria; standardized contributor agreements that describe responsibilities and acceptable use; and working groups for security, data stewardship, and user support. The NAIRR-OC solicitation (NSF 25-546) lays out expectations for a single operations center to coordinate portal development, operations, outreach and the transition to a sustainable model. Technical implementation emphasizes open metadata, API-based integration where feasible, and layered access controls to combine openness with the ability to protect sensitive datasets and adhere to export-control, privacy and contractual restrictions.

Monitoring and Evaluation

Monitoring occurs at several levels: programmatic oversight by the NSF PMO and agency partners; award-level milestones and performance reviews (including a 24-month performance review for NAIRR-OC awardees); technical monitoring of compute and data usage; and community-driven evaluation via workshops, user surveys and published case studies. Evaluation metrics include number and diversity of users served, domains supported, educational outreach outcomes, reproducibility and research outputs, security incidents, and time-to-access for resources. The pilot intentionally collects operational data to inform policies for allocation fairness, security controls, data stewardship, and sustainability planning. External peer review and NIST-aligned testing frameworks help validate testing and evaluation approaches; findings feed iterative adjustments to portal features, allocation policy, and security posture.

Penalties, Liability, and Appeals

Enforcement during the pilot relies primarily on NSF award terms, partner contributor agreements, and standard federal grants compliance mechanisms. Remedies for non-compliance include corrective action plans, suspension or termination of awards, revocation of NAIRR access allocations, requirement to return funds, reporting to federal audit offices, and potential suspension or debarment consistent with federal grant regulations. Civil or criminal liability arising from misuse of data, breaches of data-use agreements, or violations of law remain subject to existing statutory frameworks. The pilot establishes appeal pathways for allocation decisions (administrative review by NAIRR PMO or NAIRR-OC), and dispute resolution processes for partner disputes, typically mediated through contractual terms or NSF oversight procedures.

Relationship to Other Instruments

The NAIRR pilot implements recommendations from the National Artificial Intelligence Research Resource Task Force final report and is aligned with the National AI Initiative Act of 2020. It was also shaped by executive guidance on AI safety and federal coordination (the October 2023 Executive Order directing NAIRR pilot launch) and complements agency-specific AI programs (NIH, DOE, DARPA), NIST standards work on AI evaluation, and the White House AI policies and strategic plans. The pilot's data stewardship and testing practices intersect with FDA's device evaluation guidance for AI in medical devices and with interagency reporting and risk management efforts directed at frontier models. While the pilot itself does not create standalone regulation, it operationalizes interagency policy priorities and helps define standards that may inform future regulation, procurement, and research oversight.

International Alignment

The NAIRR pilot pursues alignment with international norms in AI research infrastructure and testing while prioritizing U.S. research leadership and security. The pilot’s focus on interoperable metadata, reproducible evaluation, and open datasets is consistent with international research infrastructure principles, and pilot partners coordinate with global research institutions on standards and benchmarking. Security and export-control considerations limit cross-border dissemination of certain datasets and compute capabilities; however, the pilot supports international collaboration where legally permissible and beneficial to science. Lessons from NAIRR pilot governance and test/evaluation frameworks are intended to inform international standard-setting bodies and bilateral/multilateral research partnerships.

Implementation Timeline

Sources and References