Future of Artificial Intelligence Innovation Act of 2026 — S. 3952 bill text (introduced February 26, 2026)

S. 3952 is a Senate bill of the 119th Congress introduced on February 26, 2026 by Sen. Todd Young (R-IN) "for himself, Ms. Cantwell, Mrs. Blackburn, and Mr. Hickenlooper" and referred to the Committee on Commerce, Science, and Transportation. Its short title is the "Future of Artificial Intelligence Innovation Act of 2026." The text runs to three titles and fifteen sections. Its principal provision, section 101, would insert a new section 22B into the National Institute of Standards and Technology Act establishing the Center for Artificial Intelligence Standards and Innovation within NIST; the remaining sections set up a joint NIST–Department of Energy testbed program, international standards coalitions, a Government Accountability Office report on regulatory impediments, a public-dataset priorities list, federal AI prize challenges, and research-security and personnel provisions. The long title describes the bill as one "[t]o establish artificial intelligence standards, metrics, and evaluation tools, to support artificial intelligence research, development, and capacity building activities, to promote innovation in the artificial intelligence industry by ensuring companies of all sizes can succeed and thrive, and for other purposes."

The page summarizes the text captured from GovInfo (BILLS-119s3952is). Provisions are described as drafted; the bill has not been marked up or enacted.

Provenance and versions

Legislative status. Congress.gov records two actions: referral to Commerce on February 26, 2026, and "Star Print ordered on the bill" on March 19, 2026. The three cosponsors, Sens. Maria Cantwell (D-WA), Marsha Blackburn (R-TN) and John W. Hickenlooper (D-CO), are all original cosponsors dated February 26, 2026; none has been added since. The bill has no related bills, no CBO cost estimate and no CRS summary, and its status is "Introduced" (Source: congress.gov). It was not among the nine bills on the committee's September 16, 2026 executive session, and the committee's meetings list retrieved on September 17, 2026 showed no executive session scheduled for September 23 (Source: commerce.senate.gov; commerce.senate.gov).

Introduced text and star print. Congress.gov lists a single text version, "Introduced in Senate." The GovInfo package for that version is itself the first star print: its XML carries star-print="first-star-print", its metadata record was created on March 20, 2026 with a star-print number of 1, and its PDF is stamped "Star Print." No earlier printing of the introduced bill is published on either site, so the public record does not show what the star print changed. The Senate Commerce Committee's minority press release of February 26, 2026 linked a Legislative Counsel draft carrying the same drafting identifier (BAG26179) and the sponsorship line "Mr. YOUNG (for himself and Ms. CANTWELL)" (Source: commerce.senate.gov). A word-by-word comparison of that draft with the star print finds the operative text identical except for the capitalization of "Federal Government" in two places, "decision-making" printed as "decision making" in section 111(c), and a closing quotation mark after "30" in section 302(a). The sponsorship line is the only difference of substance between the two documents.

Relation to S. 4178 (118th Congress). The committee release describes the February 2026 measure as a reintroduction, and says it builds on Cantwell and Young's earlier FUTURE of AI Act, which it credits with creating the National AI Advisory Committee (Source: commerce.senate.gov). The predecessor is S. 4178, the Future of Artificial Intelligence Innovation Act of 2024, introduced April 18, 2024 by Sen. Cantwell with Sens. Young, Hickenlooper and Blackburn; Sens. Wicker, Luján, Sinema, Rounds and Schumer joined later. The committee ordered it reported with an amendment in the nature of a substitute on July 31, 2024, reported it without written report on December 18, 2024, and it was placed on the Senate calendar as Calendar No. 725 (Source: congress.gov). S. 3952 follows the section structure of that reported substitute, with these differences in the text:

The S. 4178 substitute's section 304 would have directed a presidential technology directive prohibiting federal agency policies that promote listed concepts, including that AI systems "should be designed in an equitable way that prevents disparate impacts" (Source: govinfo.gov). S. 3952 carries no counterpart. The 2024 bill as introduced also contained a section 112 on bilateral and multilateral research collaborations and, among the institute's functions, voluntary guidelines on "transparency documentation of artificial intelligence datasets and artificial intelligence models," both of which the reported substitute dropped (Source: govinfo.gov); neither appears in S. 3952.

Title I — Voluntary standards, metrics, evaluation tools, testbeds, and international cooperation

The full title heading is "Voluntary Artificial Intelligence Standards, Metrics, Evaluation Tools, Testbeds, and International Cooperation." Section 2, preceding it, states a sense of Congress "that policies affecting artificial intelligence should maximize the potential, development, and use of artificial intelligence to benefit all private and public stakeholders."

Section 100 — Definitions

For Title I, "artificial intelligence" takes its meaning from section 5002 of the National Artificial Intelligence Initiative Act of 2020 (15 U.S.C. 9401). An artificial intelligence system is "an engineered or machine-based system that (A) can, for a given set of objectives, generate outputs such as predictions, recommendations, or decisions influencing real or virtual environments; and (B) is designed to operate with varying levels of autonomy." An artificial intelligence model is a component of such a system derived using mathematical, computational, statistical or machine-learning techniques and used to produce outputs from inputs. A foundation model is "an artificial intelligence model trained on broad data at scale and is adaptable to a wide range of downstream tasks." A testbed is "a facility or mechanism, virtual or otherwise, equipped for conducting rigorous, transparent, and replicable testing of tools and technologies, including artificial intelligence systems, to help evaluate the functionality, trustworthiness, usability, and performance of those tools or technologies." Critical infrastructure, Federal laboratory and National Laboratory take existing statutory definitions (42 U.S.C. 5195c(e), 15 U.S.C. 3703 and 42 U.S.C. 15801).

Section 101 — Center for Artificial Intelligence Standards and Innovation

Section 101 amends the NIST Act (15 U.S.C. 271 et seq.) by inserting a new section 22B after section 22A (15 U.S.C. 278h-1). Section 22A is the provision a House discussion draft of August 2026 would amend to condition the Director's pre-deployment evaluation agreements (Self-Improving AI Monitoring Act — discussion draft bill text (August 31, 2026)); section 22B would sit beside it.

Establishment and mission (§ 22B(b)). "Not later than 90 days after the date of the enactment of the Future of Artificial Intelligence Innovation Act of 2026, the Director shall establish a center on artificial intelligence within the Institute," to be known as the Center for Artificial Intelligence Standards and Innovation. Its mission "is to assist the private sector and agencies in developing voluntary best practices for the robust assessment of artificial intelligence systems, which may be contributed to or inform the work on such practices in standards development organizations." The section does not create a separate director for the Center, although section 104(d) of the bill refers to "the Director of the Center for Artificial Intelligence Standards and Innovation."

Definitions (§ 22B(a)). Artificial intelligence red-teaming "means structured adversarial testing efforts of an artificial intelligence system." Artificial intelligence blue-teaming means "an effort to conduct operational vulnerability evaluations and provide mitigation techniques to entities who have a need for an independent technical review of the security posture of an artificial intelligence system." Synthetic content is information "that has been significantly modified or generated by algorithms, including by an artificial intelligence system." Watermarking is "the act of embedding provenance and authenticity information that is intended to be difficult to remove" into AI outputs or original content, for verifying and maintaining "the authenticity, integrity, and reliability of the output or the identity or characteristics of its provenance, modifications, or conveyance." Agency, artificial intelligence, Federal laboratory, foundation model and testbed are also defined, the last two in the same terms as section 100.

Functions (§ 22B(c)). The Center carries out its functions "in coordination with the laboratories of the Institute." The seventeen enumerated functions are:

• (A) "Using publicly available or voluntarily provided information, assessing artificial intelligence systems and developing guidelines and best practices to measure and improve the secure development, deployment, and use of artificial intelligence technology."
• (B) Supporting AI red-teaming, sharing best practices, and coordinating on building testbeds and test environments "with allies and international partners of the United States."
• (C) Developing and publishing physical and cybersecurity tools, methodologies, best practices and voluntary guidelines to help persons who maintain systems used to create or train AI models discover and mitigate vulnerabilities and attacks, "including manipulation through data poisoning, including those that may be exploited by foreign adversaries."
• (D) Establishing AI blue-teaming capabilities and partnering with industry on the reliability of AI systems.
• (E) Developing tools, methodologies, best practices and voluntary guidelines for detecting synthetic content, authenticating content and tracking its provenance, labeling original and synthetic content "such as by watermarking," and evaluating detection and labeling software.
• (F) Coordinating or developing metrics and methodologies for testing AI systems, including cataloging existing metrics and evaluation methodologies used in industry and academia, testing the efficacy of existing metrics and evaluations, and documenting tools that assess reliability, accuracy and robustness.
• (G) Coordinating with counterpart international institutions, partners and allies "to support global interoperability in the development of research and testing of standards."
• (H) "Producing resources for Federal agencies to conduct their own evaluations of artificial intelligence systems to best fulfill their missions."
• (I) Convening semiannual meetings with federal agencies and the private sector to share information on building AI evaluations and to accelerate national standards for AI systems in sectors "including, biotechnology, agriculture, and health care."
• (J) Examining safeguards and best practices to secure AI systems from cyber attacks.
• (K) Examining safeguards and best practices "to protect against unintended use of artificial intelligence for the purpose of developing chemical, biological, radiological, nuclear, and energy-security threats or hazards."
• (L) Providing, in consultation with the Secretary of Homeland Security and the Director of the "Cybersecurity and Information Security Agency" (so named in the text), a toolkit for anticipating, responding to and recovering from cybersecurity incidents involving AI systems, which may include guidance on known AI-specific vulnerabilities.
• (M) Developing and curating, with the Secretary of Labor, a list of high-priority occupations for training for the advancement and deployment of AI.
• (N) Developing best practices on minimum data quality standards for biological, material science, chemical, physical and other scientific data in AI model training.
• (O) Examining, with other relevant agencies, "the vulnerabilities in the supply chain of hardware, including semiconductors and microelectronics, that are critical to enabling the development and deployment of artificial intelligence."
• (P) Examining ways AI may be used by the federal government in combating fraud and other unfair or deceptive practices.
• (Q) "Identify proven, scalable, and interoperable techniques and metrics to promote the development of artificial intelligence."

Consortium (§ 22B(d)). Within 180 days of enactment the Director must establish the "Center for Artificial Intelligence Standards and Innovation Consortium," drawn from academic or research communities, Federal laboratories, private industry "including companies of all sizes with different roles in the use of artificial intelligence systems, including developers, deployers, evaluators, users," and civil society. The Director consults it at least quarterly and reports annually to the Senate Commerce Committee and the House Science, Space, and Technology Committee on its members' contributions.

Voluntary testing standards (§ 22B(e)). In carrying out the Center's functions, the Director "shall support and contribute to the development of voluntary, consensus-based technical standards for testing artificial intelligence system components," addressing as the Director considers appropriate: physical infrastructure, including cloud infrastructure, for training or developing models and for operating systems; training data; data for evaluating the functionality and trustworthiness of trained models and systems; trained or partially trained models "and any resulting software systems or products"; and human-in-the-loop testing.

Disclosure and access (§ 22B(f)). Confidential content, "as deemed confidential by the contributing private sector person," is exempt from public disclosure under 5 U.S.C. 552(b)(3). Access to a contributor's voluntarily provided confidential content is limited to that person and the Center. The Director may release aggregated, deidentified information to contributing companies, the public and other agencies.

Rule of construction (§ 22B(g)). The subsection reads in full:

"Nothing in this section shall be construed to provide the Director any enforcement authority that was not in effect on the day before the date of the enactment of the Future of Artificial Intelligence Innovation Act of 2026."

The clause is scoped to "this section," that is, to section 22B, and it is the same clause S. 4178 carried, with the year changed. Nothing in section 22B requires any developer to submit a model for assessment: function (A) is framed around "publicly available or voluntarily provided information," and the access rules in subsection (f) govern content that is "voluntarily provided."

Foreign-government control (§ 22B(h)). An entity "under the ownership, control, or influence of the government of a covered nation" may not access any of the Center's resources. Covered nation takes its meaning from 10 U.S.C. 4872. The definition of ownership, control or influence covers a covered-nation government's power to direct an entity's management or operations in a way that could result in unauthorized access to classified information or adversely affect a contract requiring such access, exercised directly or indirectly, through securities ownership or contract, through board appointments, or "prospectively (such as by not currently exercising the power, but could)." The Director, with the heads of relevant agencies, sets criteria for identifying such entities.

Section 102 — Interagency coordination and testbed program

Section 102 is a free-standing provision, not an amendment to the NIST Act. In it "Director" means the Director of the National Science Foundation, "Under Secretary" the Under Secretary of Commerce for Standards and Technology, and "Secretary" the Secretary of Energy; the appropriate committees are Senate Commerce, Senate Energy and Natural Resources, and House Science, Space, and Technology.

The program (§ 102(b)). Within one year of enactment, the Under Secretary and the Secretary, in coordination with the NSF Director, "shall jointly establish a testbed program to encourage collaboration and support partnerships between the National Laboratories, Federal laboratories, the National Institute of Standards and Technology, the National Artificial Intelligence Research Resource pilot program established by the Director, or any successor program, and public and private sector entities, including companies of all sizes, to conduct tests, evaluations, and security or vulnerability risk assessments, and to support research and development, of artificial intelligence systems, including measurement methodologies developed by the Institute, in order to develop standards and encourage development of a third-party ecosystem."

Activities (§ 102(c)). The Under Secretary and the Secretary may use the advanced computing resources, testbeds and expertise of the National Laboratories, Federal laboratories, NIST, NSF and private entities "to run tests and evaluations on the capabilities and limitations of artificial intelligence systems." They shall use existing solutions to the maximum extent practicable; develop automated and reproducible tests where practicable; assess the computational resources needed for testing and research ways to minimize them; where practicable develop tests for high-, medium- and low-computational intensity; and organize a hackathon to test AI security risks and vulnerabilities. Paragraph (7) provides that they "shall prioritize assessments by identifying security vulnerabilities of artificial intelligence systems, including the establishment of and utilization of existing classified testbeds, at the National Laboratories if necessary," with respect to autonomous offensive cyber capabilities; cybersecurity vulnerabilities in the AI software ecosystem "and beyond"; chemical, biological, radiological, nuclear, critical infrastructure and energy-security threats or hazards; and other capabilities they determine necessary.

Biological models (§ 102(d)). The officials must consider the applicability of tests, evaluations and risk assessments to AI systems "trained using primarily biological sequence data that could be used to enhance an artificial intelligence system's ability to contribute to the creation of a pandemic or biological weapon, including those systems used for gene synthesis."

Voluntary foundation models test program (§ 102(i)). Within the program, the Under Secretary and the Secretary "shall, jointly carry out a test program to provide vendors of foundation models, as well as vendors of artificial intelligence virtual agents and robots that incorporate foundation models, the opportunity to voluntarily test foundation models across a range of modalities, such as models that ingest and output text, images, audio, video, software code, and mixed modalities."

Other terms. The officials develop metrics for the program's effect on collaboration and on public and private integration and use of AI (§ 102(e)), may run the program through a program already in effect (§ 102(f)), and within three years of its start must evaluate it and report findings to the appropriate committees (§ 102(g)). They consult, as they consider appropriate, industry "including private artificial intelligence laboratories, companies of all sizes, and representatives from the United States financial sector," academia and civil society (§ 102(h)). Confidential content is handled as under section 22B, except that access is limited to the contributor "and the Institute" (§ 102(j)). A separate rule of construction provides that nothing in the section "shall be construed to require a person to disclose any information," including trade secrets or other protected intellectual property, confidential business information, or privileged information (§ 102(k)). The programs under subsections (b) and (i), and the section's requirements, terminate seven years after enactment (§ 102(l)).

Section 103 — NIST and DOE materials testbed

Section 103 is headed "National Institute of Standards and Technology and Department of Energy testbed to identify, test, and synthesize new materials," but its operative text does not establish a second testbed. Subsection (a) provides that the Secretary of Commerce, acting through the Under Secretary, and the Secretary of Energy "may use the program established under section 102(b) to advance materials science and energy storage and optimization and to support advanced manufacturing for the benefit of the United States economy through the use of artificial intelligence, autonomous laboratories, and artificial intelligence integrated with emerging technologies, such as quantum hybrid computing and robotics." The two Secretaries must ensure the technologies accelerated are supported by advanced algorithms and models, uncertainty quantification, and software and workforce development tools producing benchmark data, model comparison tools and best-practice guides (§ 103(b)); must enter public-private partnerships in consultation with industry, civil society and academia as they jointly determine appropriate (§ 103(c)); and may draw on the Manufacturing USA Program, the Hollings Manufacturing Extension Partnership, the National Laboratories, Federal laboratories, the private sector and the section 102(b) program (§ 103(d)). The phrase "identify, test, and synthesize new materials" appears only in the heading. Section 102 thus covers AI evaluation and security testing generally and carries the voluntary foundation-model program; section 103 is a permissive authority to apply that program to materials science, energy storage and advanced manufacturing.

Section 104 — Coordination, reimbursement and savings

The Secretary of Commerce must ensure no duplication between activities under Subtitle A and those of Department of Energy research entities, including the National Laboratories and the Advanced Scientific Computing Research program, and of relevant industries. Energy Department or National Laboratory resources provided to NSF, NIST or other entities under the subtitle are provided "on a reimbursable basis" and "pursuant to a reimbursable agreement"; "the Secretary" may waive that requirement if it determines a waiver "is necessary or appropriate to carry out the missions of the Department of Commerce" (the section does not define "Secretary"). The savings clause provides that nothing in the subtitle modifies section 5501 of the National Artificial Intelligence Initiative Act of 2020 (15 U.S.C. 9461) or allows the Secretary of Commerce, the Under Secretary, the Director of the Center, or the NSF Director "to use monetary resources of the Department of Energy or any National Laboratory."

Section 105 — Progress report

Within one year of enactment the Under Secretary, coordinating with the Secretaries of Commerce and Energy, reports to Congress on implementing sections 102 and 103, including the reimbursable agreements, statements of work, project schedules and deliverables for the testbed program, and the total amount of reimbursable agreements under section 104(b).

Section 111 — International standards coalitions

The Under Secretary and the Secretary of Energy jointly lead information exchange among federal agencies, and communication to the US private sector and "like-minded governments of foreign countries," to ensure effective federal engagement in international technical standards for AI (§ 111(a)). The Under Secretary "shall seek to form alliances or coalitions with like-minded governments" to support private sector-led standards development; to encourage adoption of US-developed technical standards by international standards organizations "and to advocate for international approaches to governance of artificial intelligence that promote innovation and counter influence from foreign adversaries"; to facilitate international collaboration on AI research and development, "including data sharing, expertise, and resources"; to build government-to-government infrastructure using existing bilateral and multilateral agreements where practicable; to identify best practices for AI-model cybersecurity; and to coordinate with the Secretary of State, the National Security Council and NSF on complementary technology protection measures, "including in basic research and higher education" (§ 111(b)).

Participation is governed by "technology trust criteria" set jointly by the Secretaries of Commerce, Energy and State and "the Director," requiring a high level of scientific and technological advancement and support for the principles of the WTO Technical Barriers to Trade Committee decision on international standards (G/TBT/1), "such as transparency, openness, and consensus-based decision making" (§ 111(c)). Partner countries must have sufficient intellectual property protections, safety standards and risk management approaches, and must develop and coordinate research security measures, export controls and IP protections (§ 111(e)). The People's Republic of China is ineligible until the US Trade Representative determines, in the report required by section 421 of the U.S.-China Relations Act of 2000 (22 U.S.C. 6951), that China "has come into compliance with the commitments it made in connection with its accession to the World Trade Organization"; after such a determination the officials must report to Congress with a national-security justification, a risk assessment including technology or IP transfer, a plan for addressing human rights concerns, and an assessment of their ability to monitor China's commitments (§ 111(f)). Nothing in the section prohibits a person from participating in an international standards body or constrains "separate engagement with emerging economies on artificial intelligence" (§ 111(g)). "Director" is not defined within section 111.

Section 121 — GAO report on regulatory impediments

Under Subtitle C, "Identifying Regulatory Barriers to Innovation," the Comptroller General must report to Congress within one year "on regulatory impediments to innovation in artificial intelligence systems." The report covers significant examples of federal statutes and regulations that directly affect AI innovation, "including the ability of companies of all sizes to compete," accounting for the effect of voluntary standards and best practices developed with federal contributions; progress in government adoption of AI; examples of AI helping agencies deliver services, "including towards combating fraud"; "Examples of Federal laws and regulations relating to infrastructure and energy that unduly burden artificial intelligence systems"; and recommendations for legislative or administrative action "to increase the rate of innovation in artificial intelligence systems." See Government Accountability Office (GAO).

Title II — Research, development and capacity building

Section 201 — Public data for AI systems (new NAIIA § 5103A)

Section 201 adds section 5103A to the National Artificial Intelligence Initiative Act of 2020. The Director of the Office of Science and Technology Policy, acting through the National Science and Technology Council and the Interagency Committee and consulting the Advisory Committee on Data for Evidence Building, must develop "a list of priorities for Federal investment in creating or improving curated, publicly available Federal Government data for training and evaluating artificial intelligence systems and identify an appropriate location to host curated datasets." The list must prioritize data advancing "novel artificial intelligence systems in the public interest," datasets resulting from federally funded scientific research, and datasets unlikely to receive sufficient private support; "In carrying out subparagraph (A)(ii), the Director shall identify 20 datasets to be prioritized." Twelve considerations follow, among them the societal challenges and key technology focus areas of 42 U.S.C. 19107; sectors including agriculture, health care, transportation, manufacturing, biotechnology, communications and weather services; "Inclusion of data representative of the entire population of the United States"; national security threats from releasing datasets; the 2016 National AI R&D Strategic Plan "and the priorities listed in Winning the Race, America's AI Action Plan, dated July 2025" (America's AI Action Plan); data already available to the National AI Research Resource pilot; and requirements for federally funded researchers to disclose nonproprietary, nonsensitive datasets used by AI models. The list is subject to public comment.

The Interagency Committee may establish or leverage initiatives, including public-private partnerships, to create the datasets, must keep the work consistent with privacy law including the National Secure Data Service standards, and "shall ensure that no data sharing is permitted with any country" the Secretary of Commerce, with the Secretaries of Defense, State and Energy and the Director of National Intelligence, determines to be engaged in conduct detrimental to US national security or foreign policy. Datasets created or improved by an agency go into that agency's comprehensive data inventory under 44 U.S.C. 3511(a), and may be made available to the NAIRR pilot and Energy Department programs "in accordance with Executive Order 14110 ... or any successor program" (Executive Order 14110 (RESCINDED), rescinded in January 2025). OSTP reports within a year on best practices, lessons learned, principles for AI-ready data, data-standards recommendations, incentives for federally supported researchers to release high-quality datasets while protecting personally identifiable information and national and economic security, and "secure compute environments at the National Science Foundation to enable secure artificial intelligence use cases for controlled access to restricted Federal data." Nothing in the section requires disclosure of trade secrets, confidential business information or privileged information, or prohibits an agency from withholding information from a public dataset.

Section 202 — Federal grand challenges in AI (new NAIIA § 5107)

Section 202 adds section 5107. Within a year of enactment the OSTP Director, through the National Science and Technology Council, and the Interagency Committee "may establish a program to award prizes" under section 24 of the Stevenson-Wydler Technology Innovation Act of 1980 (15 U.S.C. 3719), to expedite AI development in the United States and to stimulate research, development and commercialization solving "specific, well-defined, and measurable challenges." They "shall identify, and annually review and update," a list of priorities, consulting industry, civil society and academia. The initial list "may include" fourteen priorities: microelectronics engineering and applied research, including integration with neuromorphic and quantum computing; long-term advances through next-generation algorithm design and compute capability, generative and adaptive AI for design, photonics-based microprocessors and optical networks, the chemistry and physics of new materials, biotechnology "such as modeling a single cell," energy efficiency, "cryptographically secure content provenance information," and "safety and controls for artificial intelligence applications"; "explainability and mechanistic interpretability"; fundamental understanding of AI capabilities in domains such as "Interpretability," "Control," "Steerability" and "Robustness against foreign adversaries"; advanced manufacturing; AI in maritime vessel navigation, propulsion and fuels; border security, "including solutions relevant to the detection of fentanyl, illicit contraband, and other illegal activities"; AI for science; cybersecurity for AI-related intellectual property; modernizing obsolete code in government agencies and critical infrastructure; reducing energy consumption of AI models; preventing "misuse of artificial intelligence systems for malicious purposes"; wireless communications; and AI, robotics and automation for physical and cloud laboratories.

The OSTP Director, the NIST Director, the DARPA Director, other agency heads OSTP considers relevant, and the National Artificial Intelligence Advisory Committee each identify and select grand challenges; OSTP seeks public input and publishes problem statements, targets, success metrics and validation protocols on Challenge.gov. "Subject to the availability of amounts appropriated for this purpose," the Secretary of Commerce, the Secretary of Transportation and the NSF Director may establish prize competitions, challenge-based acquisitions or other research investments aligned with the priorities. A private prize recipient must be incorporated in and maintain a primary place of business in the United States; an individual must be a citizen or permanent resident; judges may come from the private sector; each competition must benefit US economic competitiveness and industry, use US advanced manufacturing where it involves manufacturing, and conduct all research in the United States. Awarding agencies notify the Commerce and Science committees within 60 days of a prize and publish a biennial report. The section terminates five years after enactment.

Section 202(b) directs two GAO studies: an initial study of federal prize competitions generally, assessing among other things whether competitions shortened development timelines, were cost effective, or outperformed grants or contracts, with a briefing and report within 540 days; and an interim study of the AI prize challenges themselves.

Title III — Research security and other matters

• Section 301 requires activities under the Act to be carried out in accordance with subtitle D of title VI of the Research and Development, Competition, and Innovation Act (42 U.S.C. 19231 et seq., part of Public Law 117-167) and section 223 of the FY2021 National Defense Authorization Act (42 U.S.C. 6605).
• Section 302 amends section 6 of the NIST Act (15 U.S.C. 275) by striking "15" and inserting "30" in the second sentence of subsection (b), and replaces the subsection (c) sunset with "shall expire on December 30, 2035." In current law the second sentence of section 275(b) reads "The Director shall appoint not more than 15 personnel under this section," and the hiring authority expires five years after August 9, 2022 (Source: law.cornell.edu).
• Section 303 requires a temporary fellow (a fellow, contractor, consultant or other non-federal employee working for an agency) and the agency head to sign, before the fellow performs work under the Act relating to AI and other critical and emerging technologies, "a certification that the temporary fellow will not perform any inherently governmental functions," with copies to the OMB Director and the chairs and ranking members of the committees of jurisdiction within 30 days. Each agency's inspector general must audit annually the number of temporary fellows, "the entities paying any temporary fellow for their work for the agency," their work, the hiring authorities used, and compliance.

Drafting notes in the text

Several cross-references in the introduced text do not resolve as written. Section 202(b)(2) directs a study of prize challenges "implemented under section 5108 of the of the National Artificial Intelligence Initiative Act of 2020, as added by subsection (a)," while subsection (a) inserts section 5107; its report is due 180 days after enactment while its briefing is due a year after the study is completed. Both GAO briefings go to "the Committee on Science, Space, and Technology and the Committee on Energy and Natural Resources of the Senate and the Committee on Energy and Commerce of the House of Representatives," although the Science, Space, and Technology Committee is a House committee. Section 121(b)(5) refers to findings "with respect to paragraphs (1) through (5)" from within paragraph (5). New section 5107(a)(1) refers to eligibility "pursuant to subsection (e)," which contains reports; eligibility is in subsection (d). Section 303(c)(1)(E) refers to "section (b)." The 5108 reference and the committee list also appear in the reported text of S. 4178 (Source: govinfo.gov).

Sponsors' description and reception

The Commerce Committee minority release of February 26, 2026 quotes Cantwell: "This legislation brings together private sector and government experts to develop voluntary standards for AI, create new assessment tools, and conduct testing that will ensure the United States leads in AI-driven innovation and competitiveness for decades to come." Young is quoted: "One of my top priorities for federal AI policy is to ensure these technologies are developed in a manner that reflects American values and supports U.S. innovation." The release states that provisions "were drafted based on recommendations from NAIAC reports" and that the bill "builds on CHIPS and Science Act investments" (Source: commerce.senate.gov).

The same release carries supporting statements from the Software & Information Industry Association (Paul Lekas, who says the UK AI Security Institute "has 10x more staff and 10x more funding" than CAISI), the Center for Data Innovation at ITIF (Hodan Omaar), the Alliance for Secure AI (Brendan Steinhauser), Americans for Responsible Innovation (Brad Carson), TechNet (Linda Moore), SeedAI (Austin Carson), the Foundation for American Innovation (Samuel Hammond, who says the bill "aligns CAISI and other federal agencies' missions to the White House's AI Action Plan and Genesis Mission") and the Business Software Alliance (Craig Albright). FedScoop reported that the previous version had advanced out of committee by voice vote in 2024 (Source: fedscoop.com).

Some descriptions of the bill do not match its text:

• The committee release's item 1 heading reads "Authorizes the NIST AI Safety Institute to Develop AI Standards" and says the body "will develop performance benchmarks, evaluations and clear transparency documentation standards." Section 22B contains no reference to transparency documentation; that phrase appeared in S. 4178 as introduced in 2024.
• The release's item 5 says the bill "creates a multilateral research collaboration between scientific and academic institutions across the world." S. 3952 has no such section. The nearest provision is the coalition objective in section 111(b)(3); the separate collaboration requirement was section 112 of S. 4178 as introduced and was dropped from the 2024 reported substitute. MeriTalk's March 3, 2026 account repeats the description (Source: meritalk.com).
• MeriTalk also reported that the bill would "mandate federal science agencies to make at least 20 curated datasets publicly available." The text directs the OSTP Director to "identify 20 datasets to be prioritized" in a list of investment priorities.
• A Washington Post newsletter of September 14, 2026 described the bill as creating "new public-private testbeds hosted by the Department of Energy, the National Science Foundation and the National Institute for Standards and Technology to evaluate AI models" (Source: washingtonpost.com). In the text, one program is established jointly by the Under Secretary of Commerce for Standards and Technology and the Secretary of Energy "in coordination with" the NSF Director, and draws on National Laboratory, Federal laboratory, NIST, NSF and NAIRR resources.

Relationships

• instance-of: Future of Artificial Intelligence Innovation Act of 2026 (S. 3952) — the legislation page for this bill
• supports: NIST CAISI (Center for AI Standards and Innovation) — section 101 would give the Center a statutory basis as NIST Act § 22B
• related: National Institute of Standards and Technology (NIST) — host agency; section 302 raises its critical-technical-expert hiring cap
• related: Department of Energy (DOE), NSF — National Science Foundation (AI research funder) — co-lead and coordinating agency for the section 102 testbed program
• related: Office of Science and Technology Policy (OSTP) — dataset priorities (§ 5103A) and grand challenges (§ 5107)
• related: Government Accountability Office (GAO) — section 121 report and section 202(b) prize-competition studies
• related: National AI Advisory Committee (NAIAC) — the release says provisions were drafted from NAIAC recommendations; NAIAC selects grand challenges under § 5107(b)(3)
• related: Todd Young, Maria Cantwell, Marsha Blackburn, John Hickenlooper — sponsor and original cosponsors
• related: Self-Improving AI Monitoring Act — discussion draft bill text (August 31, 2026) — would amend NIST Act § 22A, beside the § 22B this bill would add
• related: Frontier Act / Great American AI Act (Obernolte–Trahan) — House vehicle that would also codify the Center
• related: S. 2938 — Artificial Intelligence Risk Evaluation Act of 2025 (bill text) — a Senate bill placing a mandatory evaluation program at the Department of Energy, where S. 3952's DOE-linked testing is voluntary
• related: AI Pre-Release Vetting, Jailbreaking and Red Teaming, AI Content Provenance, CBRN Uplift, AI and Cybersecurity
• related: America's AI Action Plan, Executive Order 14110 (RESCINDED) — both named in new § 5103A