United States - AI Partnerships RFI

Overview

The Department of Energy (DOE), Office of Science, published a Request for Information (RFI) titled “Partnerships for Transformational Artificial Intelligence Models” in the Federal Register on December 5, 2025. The RFI implements a direction in Section 50404 of Public Law 119-21 and requests stakeholder input on establishing a public–private consortium to curate DOE scientific data from the National Laboratory complex for use in AI models and on approaches to develop self‑improving AI models for science and engineering. Responses are due January 14, 2026 and must be submitted electronically to [email protected]. The complete RFI package is posted on SAM.gov.

Definitions

Key statutory and programmatic terms used in the RFI include: “American science cloud” as defined in Section 50404 of Public Law 119‑21 (a system of U.S. government, academic, and private‑sector programs and infrastructures utilizing cloud computing to support research); “artificial intelligence” as defined by the National Artificial Intelligence Initiative Act of 2020; “self‑improving AI models” which DOE uses to describe models that incorporate iterative learning and model update processes informed by new data and scientific feedback; and “curation” meaning cataloging, metadata enrichment, quality control, and access governance for scientific datasets across National Laboratories.

Governance and Institutional Framework

DOE seeks input on governance structures for the proposed public‑private consortium, including membership models (e.g., multi‑stakeholder boards, technical working groups), decision‑making authorities, intellectual property and licensing frameworks, contracting and financial models (grants, cooperative agreements, cost‑sharing, subscription models), and the roles of National Laboratories. The RFI asks stakeholders to address how to coordinate across federal programs and comply with applicable laws and regulations (e.g., data control statutes, security clearances, procurement rules). DOE highlights that models and curated data will be provided through a system of U.S. government, academic, and private infrastructure (the American science cloud), and asks how to structure service‑level agreements, data stewardship responsibilities, and long‑term sustainability plans for the consortium and the underlying infrastructure. Respondents are invited to recommend governance safeguards to prevent conflicts of interest and to balance open scientific access with protection for controlled information.

Key Focus Areas

DOE’s RFI targets several technical and policy focus areas: (1) Data curation and standards — recommendations on metadata, ontologies, provenance, quality metrics, and interoperability; (2) Access and distribution models — open datasets vs. tiered/controlled access, authentication and authorization approaches, and catalog services; (3) Model development lifecycle — standards for dataset partitioning, training/validation/testing pipelines, benchmark datasets, and reproducibility practices; (4) Cloud and infrastructure integration — hybrid cloud/on‑premises strategies, data egress/cost considerations, and secure enclaves; (5) Security, compliance and privacy — classification handling, export controls, Personally Identifiable Information (PII) protections, and cyber defense; (6) Intellectual property, licensing and liability — model and dataset licensing approaches, commercialization pathways, and liability allocation; (7) Evaluation and certification — model testing, red teaming, and third‑party evaluation frameworks; and (8) Workforce and community engagement — training, user support, and mechanisms for broad scientific participation.

Implementation Framework

The RFI requests concrete proposals for how DOE should initiate and operationalize the consortium: suggested timelines, pilot projects, technical standards, and procurement/partnering approaches (e.g., cooperative agreements, CRADAs, public‑private consortia, challenge prize mechanisms). DOE asks for input on partner selection criteria (technical capacity, prior experience with sensitive datasets, cloud operations), proposed funding models and cost‑sharing arrangements, recommended legal and contracting templates, and approaches to integrate National Laboratory data holdings with commercial cloud providers while preserving federal controls. Respondents are asked to recommend phased rollouts and indicators of success, including pilot use cases in energy, materials science, climate modeling, and other domain sciences.

Monitoring and Evaluation

DOE seeks views on monitoring, evaluation, and performance metrics to judge the consortium’s effectiveness. Suggested metrics include dataset coverage and quality, model accuracy and robustness on benchmark tasks, latency and accessibility of model services, uptake by the scientific community, reproducibility rates, security incident metrics, and contributions to discovery (publications, validated results). DOE also requests input on independent evaluation mechanisms (e.g., peer review boards, external auditors, and third‑party testing), continuous monitoring for model drift and failure modes, and transparency reporting on usage and governance actions. The RFI asks how to balance transparency with security and proprietary concerns when publishing performance data.

Penalties, Liability, and Appeals

The RFI is an information‑gathering notice and does not itself impose penalties; however DOE requests feedback on contractual approaches to allocate liability among consortium participants (e.g., indemnities, insurance, limitation of liability clauses) for harms arising from misuse, model failures, or data breaches. DOE asks about appeals or dispute resolution mechanisms within consortium governance (e.g., arbitration, ombudsperson offices), and about enforcement tools to address non‑compliance with security, data handling, or access rules among partners. Respondents may propose administrative, contractual, and technical measures to mitigate risk and specify escalation pathways for incidents affecting national security‑sensitive datasets.

Relationship to Other Instruments

The RFI situates the initiative in the context of existing federal AI and data policy instruments. It references Section 50404 of Public Law 119‑21 as the statutory basis, and aligns with the definitions in the National Artificial Intelligence Initiative Act of 2020. DOE requests comment on how the consortium and the American science cloud should coordinate with other federal initiatives, agency data‑sharing platforms, National Institute of Standards and Technology (NIST) standards, and international research infrastructures. Respondents should identify potential overlaps, gaps, or conflicts with existing authorities and recommend harmonization strategies.

International Alignment

DOE requests input on international coordination and alignment while emphasizing a U.S.-centric American science cloud architecture. The RFI asks how to enable international scientific collaboration without compromising legal, export control, or national security constraints. Respondents are asked to comment on interoperable data standards, federated access models, cross‑border data transfer safeguards, and mechanisms to participate with trusted foreign partners. DOE also seeks views on how the consortium could help sustain U.S. competitiveness in AI while aligning with international research norms and bilateral or multilateral agreements.

Implementation Timeline

Sources and References