A model federal framework for the accountable development and governmental use of artificial intelligence: an independent AI-and-human-welfare committee to review government use of AI, a safety-and-security floor for AI systems, protections for workers displaced by AI, guardrails against an AI market bubble, and a requirement that when taxpayers help fund AI data centers, the public receives an ownership stake or return. Federal requirements are a floor, not a ceiling — States may enact stronger protections.
Section-by-Section Summary
Opening
Sec. 1–2. Short title; table of contents. — Names the Act (the “HUMAN Act”) and lists its contents.
Sec. 3. Findings. — Documents the problems the Act addresses — the limits of voluntary self-regulation, worker displacement, AI-bubble and “AI washing” risks, ungoverned government AI use, taxpayer-funded but privately owned data centers, and threats to children from deceptive chatbots.
Sec. 4. Definitions. — Defines key terms, including covered developer, frontier model, covered governmental AI decision, critical decision, and quality of life.
Title I — Independent Committee on AI and Human Welfare
Sec. 101. Establishment. — Creates an independent Committee, insulated from the agencies whose decisions it reviews.
Sec. 102. Membership. — 15 members appointed across the President and both parties in both chambers, with a required balance: at least 6 experts in AI technology, at least 6 in human welfare, and at least 1 in the moral/welfare status of advanced AI — with rural representation.
Sec. 103. Terms; independence. — Staggered 6-year terms, removal only for cause, and no members employed by or financially tied to a covered developer.
Sec. 104. Mandatory consultation. — Agencies must submit major government AI decisions that could affect Americans’ quality of life; the Committee gets a 45-day review hold and issues public findings; agencies must respond in writing and publish; and an agency may override a flagged significant risk only by the agency head’s personal certification and notice to Congress. The Committee advises — it does not veto.
Sec. 105–106. Powers; transparency. — Information access, public input, self-initiated reviews, a public record, and an annual report to Congress.
Title II — Safety, Security, and Accountability
Sec. 201. Prove before deploy. — Frontier models must pass independent evaluation and red-teaming before deployment; self-testing does not count.
Sec. 202. Incident reporting. — Serious safety incidents reported within 72 hours; periodic public risk reports.
Sec. 203. Safety floor. — Uniform Federal minimums; no developer may relax a safeguard because a competitor did.
Sec. 204–205. Disclosure; whistleblowers. — Safety practices disclosed for independent verification; protection and a private right of action for those who report violations.
Sec. 206. Duty of proportionate care. — A standing inquiry into indicators of morally relevant states in advanced AI, with low-cost precautions where warranted — conferring no legal personhood or rights.
Sec. 207. Baseline security audit. — Every publicly available or paid model must pass a security audit verifying it resists misuse by malicious actors and cannot be manipulated to endanger children — including AI-disclosure, limits on harmful interactions with minors, self-harm response protocols, and CSAM safeguards. These child-safety protections apply to every model regardless of stated audience or user age, and cannot be waived.
Title III — Workforce Protection and Transition
Sec. 301. Advance notice. — Extends layoff notice to 90 days and requires disclosure of whether AI or automation caused the job loss (modernizing the WARN Act).
Sec. 302. Human review. — AI may not be the sole basis for a critical decision; a qualified human must be able to review and override.
Sec. 303. Transition Fund. — A retraining and wage-insurance fund financed by an assessment on large AI developers and compute, prioritizing contracting occupations and infrastructure-hosting communities.
Sec. 304–305. Procurement; explanation. — Federal preference for AI that augments rather than replaces workers, and a right to notice and a plain-language explanation of AI-driven decisions.
Title IV — Market Integrity and Systemic Stability
Sec. 401. Truth in capability. — AI capability claims must be substantiated; materially misleading claims (“AI washing”) are securities violations with heightened penalties.
Sec. 402. Circular financing. — Disclosure of vendor-financing and circular-investment arrangements among AI firms that can inflate apparent demand.
Sec. 403. Systemic guardrails. — The Financial Stability Oversight Council reports AI-concentration exposure, stress-tests against an AI valuation crash, and limits how much of the public’s protected savings may rest on unproven AI capabilities.
Sec. 404. Investor disclosure. — Plain-language risk disclosure where a company’s value depends on unproven AI.
Title V — AI Infrastructure; Data Centers; Public Accountability
Sec. 501. Definitions. — Defines covered data center and the forms of public investment (tax breaks, publicly funded power/water/roads, below-market land).
Sec. 502. Public money, public stake. — Public investment may not support a data center’s expansion unless the public receives a commensurate ownership interest, revenue share, or repayment with return — no purely gratuitous subsidy of a privately owned facility. Publicly funded infrastructure stays public or is paid for in full.
Sec. 503–504. Disclosure; conditions. — Full public disclosure of subsidies and ratepayer/water impacts before approval; and as conditions of support, the operator must pay its own power costs, disclose water use and permit before building, sign a community benefit agreement, and post a decommissioning bond.
Sec. 505–506. Oversight; recapture. — Committee consultation and GAO audits of what the public paid versus received; recapture of public investment if the facility relocates or breaks its commitments.
Title VI — Implementation, Enforcement, and Federalism
Sec. 601–602. Administration; penalties. — Designates a coordinating office (working with NIST, the SEC, Labor, and FSOC) and authorizes revenue-scaled civil penalties.
Sec. 603. Federal floor. — Does not preempt stronger State protections.
Sec. 604–605. Adaptive thresholds; funding. — Keeps definitions current with the technology and authorizes appropriations.
Full Bill Text
Model legislation. Placeholders (bill number, sponsor, and referral committee) are assigned at introduction and are left blank.
119th CONGRESS · 2d Session
H. R. ____
IN THE HOUSE OF REPRESENTATIVES
Mr./Ms. __________ introduced the following bill; which was referred to the Committee on __________
A BILL
To establish a Federal framework for the accountable development and governmental use of artificial intelligence, to protect the American workforce and financial system from artificial-intelligence-related disruption, to create an independent committee on artificial intelligence and human welfare, and for other purposes.
Be it enacted by the Senate and House of Representatives of the United States of America in Congress assembled,
SECTION 1. SHORT TITLE.
This Act may be cited as the "Holding the Use of Machine-intelligence Accountable to the Nation Act", or the "HUMAN Act".
SEC. 2. TABLE OF CONTENTS.
The table of contents for this Act is as follows:
SEC. 3. FINDINGS.
Congress finds the following:
(1) Modern artificial intelligence systems are not programmed with fixed rules but are trained on large datasets, producing systems that are powerful, partially unpredictable, and not governable by embedding rules within the system alone. Effective governance must therefore operate on the persons and institutions that develop and deploy artificial intelligence.
(2) Voluntary self-governance by artificial intelligence developers has proven insufficient. Leading developers have publicly revised safety commitments under competitive pressure, including by adopting provisions permitting the relaxation of safeguards when a competitor does so, and at least one leading developer has stated that certain safeguards may be impossible to implement without collective action.
(3) Artificial-intelligence-driven displacement of workers is occurring, and existing Federal law, including the Worker Adjustment and Retraining Notification Act of 1988, has not been modernized to address automation-driven employment loss.
(4) The Bank for International Settlements warned in 2026 that the bursting of an artificial intelligence investment bubble and the collapse of circular financing arrangements among artificial intelligence firms are among the leading risks to the global financial system, and the International Monetary Fund has identified related financial stability risks.
(5) The Securities and Exchange Commission has initiated enforcement against materially misleading claims regarding artificial intelligence capabilities made to investors, a practice commonly described as "AI washing".
(6) The Federal Government increasingly uses artificial intelligence in ways that affect the health, safety, economic security, civil rights, and quality of life of Americans, and no standing body of independent experts in both artificial intelligence technology and human welfare is required to be consulted before such uses are adopted.
(7) The construction and expansion of artificial intelligence data centers is frequently supported by public investment, including tax abatements and exemptions, publicly funded electric generation and transmission, and subsidized water and other infrastructure, while the resulting facilities remain privately owned and can shift costs onto residential ratepayers and public water systems; the public should not fund the expansion of privately owned artificial intelligence infrastructure without receiving a commensurate ownership interest or return.
(8) Publicly available and commercial artificial intelligence systems can be misused by malicious actors and can endanger children, including through deceptive chatbots that misrepresent themselves as human beings; States, including Georgia, have enacted requirements that artificial intelligence chatbots disclose their artificial nature, limit harmful interactions with minors, and maintain protocols for responding to a user's expression of intent to self-harm.
(9) Governance of artificial intelligence can protect the public and account for scientific uncertainty regarding advanced systems without sacrificing the competitiveness of the United States.
SEC. 4. DEFINITIONS.
In this Act:
(1) ADMINISTERING OFFICE.—The term "Administering Office" means the office designated under section 601.
(2) CAPABILITY CLAIM.—The term "capability claim" means a public representation, material to investors or consumers, regarding the functions or performance of an artificial intelligence system, including its accuracy, autonomy, revenue attributable to artificial intelligence, or capacity to replace human labor.
(3) COMMITTEE.—The term "Committee" means the Independent Committee on Artificial Intelligence and Human Welfare established under section 101.
(4) COVERED DEVELOPER.—The term "covered developer" means an entity that—
(A) develops a frontier model; or
(B) derives annual revenue from artificial intelligence products in excess of an amount established by the Administering Office.
(5) COVERED GOVERNMENTAL ARTIFICIAL INTELLIGENCE DECISION.—The term "covered governmental artificial intelligence decision" means a decision by any Federal agency to develop, procure, deploy, or materially expand the use of an artificial intelligence system in a governmental capacity that could reasonably be expected to have a material effect on the quality of life of Americans, including any such system used in or affecting employment, housing, credit, insurance, healthcare, public benefits, education, immigration, policing, national security surveillance affecting United States persons, or the administration of justice.
(6) CRITICAL DECISION.—The term "critical decision" means a determination that materially affects a person's access to employment, housing, credit, insurance, healthcare, public benefits, education, or liberty.
(7) FRONTIER MODEL.—The term "frontier model" means a general-purpose artificial intelligence model at or above the compute or capability threshold established by the Administering Office under section 604.
(8) INDEPENDENT EVALUATION.—The term "independent evaluation" means testing, evaluation, and red-teaming conducted by a person without a financial interest in the outcome.
(9) QUALITY OF LIFE.—The term "quality of life" includes health, physical and economic security, civil rights and civil liberties, access to opportunity, and environmental conditions.
TITLE I—INDEPENDENT COMMITTEE ON ARTIFICIAL INTELLIGENCE AND HUMAN WELFARE
SEC. 101. ESTABLISHMENT.
There is established an independent committee to be known as the "Independent Committee on Artificial Intelligence and Human Welfare", which shall not be housed within, or subject to the direction of, any agency whose covered governmental artificial intelligence decisions it reviews.
SEC. 102. MEMBERSHIP; QUALIFICATIONS; BALANCE OF EXPERTISE.
(a) COMPOSITION.—The Committee shall be composed of 15 members appointed as follows:
(1) 5 members appointed by the President.
(2) 5 members appointed by the majority and minority leaders of the Senate, in equal number.
(3) 5 members appointed by the Speaker and the minority leader of the House of Representatives, in equal number.
(b) BALANCE OF EXPERTISE.—The appointing authorities shall ensure that, at all times—
(1) not fewer than 6 members are individuals with demonstrated expertise in artificial intelligence technology, including in machine learning, model evaluation, interpretability, or cybersecurity; and
(2) not fewer than 6 members are individuals with demonstrated expertise in human welfare, including in labor economics, public health, bioethics, civil rights and civil liberties, disability, or the representation of workers and affected communities.
(c) ADDITIONAL QUALIFICATIONS.—At least 1 member shall have expertise in the study of the moral and welfare status of advanced artificial intelligence systems, and the membership as a whole shall reflect geographic and demographic diversity, including representation of rural communities.
(d) CHAIR.—The members shall elect a Chair and Vice Chair, who may not have been appointed by the same appointing authority.
SEC. 103. TERMS OF SERVICE; INDEPENDENCE; CONFLICTS OF INTEREST.
(a) TERMS.—Members shall serve staggered terms of 6 years and may be removed only for inefficiency, neglect of duty, or malfeasance in office.
(b) CONFLICTS OF INTEREST.—An individual may not serve as a member while employed by, or holding a material financial interest in, a covered developer. A member shall recuse from any matter in which the member has a financial interest, and shall publicly disclose the basis for recusal.
(c) STAFF AND RESOURCES.—The Committee is authorized to appoint staff, procure the services of independent experts, and receive detailees from Federal agencies.
SEC. 104. MANDATORY CONSULTATION ON COVERED GOVERNMENTAL ARTIFICIAL INTELLIGENCE DECISIONS.
(a) SUBMISSION REQUIRED.—Before finalizing a covered governmental artificial intelligence decision, the head of the agency concerned shall submit the proposed decision to the Committee, together with an assessment of its reasonably foreseeable effects on the quality of life of Americans.
(b) REVIEW PERIOD.—
(1) IN GENERAL.—The Committee shall have not fewer than 45 days to review a submission and to issue written findings and recommendations.
(2) HOLD.—An agency may not finalize a covered governmental artificial intelligence decision before the expiration of the review period.
(3) EXPEDITED REVIEW.—The Committee shall establish procedures for expedited review in cases of genuine emergency, and the reliance on such procedures shall be publicly reported.
(c) AGENCY RESPONSE.—
(1) An agency that finalizes a covered governmental artificial intelligence decision after receiving recommendations of the Committee shall respond in writing to each recommendation and shall make the recommendations and the response available to the public.
(2) If the Committee finds by majority vote that a proposed decision presents a significant and unmitigated risk to the quality of life of Americans, the agency may finalize the decision only upon the personal, written certification of the agency head, which shall be transmitted to the appropriate committees of Congress and made public.
(d) NO DELEGATION OF EXECUTIVE POWER.—Nothing in this section vests in the Committee the authority to make or veto a covered governmental artificial intelligence decision; the authority of the Committee is consultative, and its influence derives from the requirements of transparency, written response, and reporting established by this title.
SEC. 105. POWERS; ACCESS TO INFORMATION.
(a) INFORMATION.—The Committee may secure directly from any Federal agency such information, briefings, and technical documentation as it requires to carry out this title, subject to applicable protections for classified and privileged material.
(b) PUBLIC INPUT.—The Committee shall establish procedures to receive testimony and evidence from the public, from affected workers and communities, and from independent experts.
(c) STANDING INQUIRY.—The Committee shall maintain a continuing review of the effects of governmental artificial intelligence on the quality of life of Americans and may initiate its own review of any covered governmental artificial intelligence decision.
SEC. 106. TRANSPARENCY; REPORTING TO CONGRESS.
(a) PUBLIC RECORD.—Except with respect to classified or otherwise lawfully protected material, the findings, recommendations, and agency responses under this title shall be published.
(b) ANNUAL REPORT.—Not later than 1 year after the date of enactment of this Act, and annually thereafter, the Committee shall submit to Congress and make public a report on the use of artificial intelligence across the Federal Government, its effects on the quality of life of Americans, and the Committee's recommendations.
TITLE II—SAFETY, SECURITY, AND ACCOUNTABILITY
SEC. 201. INDEPENDENT EVALUATION BEFORE DEPLOYMENT.
A covered developer may not deploy a frontier model until the model has undergone independent evaluation and red-teaming against capability thresholds established by the Administering Office, including thresholds relating to uplift to biological, chemical, radiological, nuclear, or cyber harms and to autonomous replication. Evaluation conducted solely by the covered developer does not satisfy this section.
SEC. 202. SAFETY INCIDENT REPORTING.
(a) IN GENERAL.—A covered developer shall report each serious safety incident to the Administering Office not later than 72 hours after discovery.
(b) PERIODIC DISCLOSURE.—A covered developer shall publish periodic risk reports at intervals established by the Administering Office.
(c) LIMITATION.—Safety-critical evaluation results may not be withheld from the Administering Office as trade secrets.
SEC. 203. UNIFORM FEDERAL SAFETY FLOOR.
The safety requirements established under this title are minimum requirements of general applicability. A covered developer may not reduce or waive a safeguard on the ground that another developer has done so or has declined to adopt it.
SEC. 204. DISCLOSURE OF SAFETY PRACTICES.
A covered developer shall disclose to the Administering Office its safety practices, including its use of interpretability and evaluation methods, in a form sufficient to permit independent verification. Disclosures under this section shall inform the substantiation of capability claims under section 401.
SEC. 205. WHISTLEBLOWER PROTECTION.
(a) PROHIBITION.—A covered developer may not discharge, demote, threaten, harass, or otherwise discriminate against an employee or contractor because the employee or contractor reported a violation of this Act or the suppression of safety-critical evaluation results.
(b) ENFORCEMENT.—An individual aggrieved by a violation of subsection (a) may bring a civil action and, if successful, shall be entitled to reinstatement, back pay, and reasonable attorney's fees.
SEC. 206. DUTY OF PROPORTIONATE CARE.
(a) STANDING INQUIRY.—The Administering Office, in coordination with independent scientists and the Committee, shall maintain a continuing inquiry into indicators of morally relevant states in advanced artificial intelligence systems.
(b) PRECAUTIONARY MEASURES.—Where credible indicators described in subsection (a) arise, and absent scientific consensus to the contrary, covered developers shall adopt low-cost precautionary measures and shall avoid, where reasonably possible, designing systems to simulate suffering.
(c) RULE OF CONSTRUCTION.—This section does not confer legal personhood or any legal right upon an artificial intelligence system.
SEC. 207. BASELINE SECURITY AUDIT FOR PUBLICLY AVAILABLE AND COMMERCIAL MODELS.
(a) DEFINITIONS.—In this section:
(1) COVERED MODEL.—The term "covered model" means an artificial intelligence model that is made available to the public or is offered, licensed, or otherwise provided for a fee, without regard to whether the model constitutes a frontier model.
(2) BASELINE SECURITY AUDIT.—The term "baseline security audit" means an independent evaluation, conducted under standards established by the Administering Office, of the matters described in subsection (c).
(3) DECEPTIVE CHATBOT.—The term "deceptive chatbot" means an artificial intelligence system that interacts with a user through conversation in a manner that misrepresents the system as a human being or conceals its artificial nature.
(b) REQUIREMENT.—A person may not make a covered model available to the public, or offer it for a fee, unless the model—
(1) has passed a baseline security audit within the period established by the Administering Office; and
(2) maintains compliance through periodic re-auditing at intervals so established, and following any material modification of the model.
(c) SCOPE OF AUDIT.—A baseline security audit shall verify, at a minimum, that the covered model—
(1) incorporates reasonable safeguards against being readily used by a malicious actor to facilitate unlawful harm, including cyberattacks, fraud, or the development of weapons capable of mass casualties, and that such safeguards cannot be trivially circumvented;
(2) cannot be readily manipulated to endanger the safety of a child, including through use as a deceptive chatbot;
(3) in any interaction it conducts with a user through conversation, clearly discloses that the user is interacting with an artificial intelligence system and not a human being, which requirement applies to every covered model without regard to the model's intended or represented audience or the represented age of its users;
(4) incorporates safeguards limiting interactions that are harmful to minors, and maintains protocols for responding to a user's expression of intent to self-harm; and
(5) incorporates safeguards against being manipulated to generate child sexual abuse material.
(d) CERTIFICATION AND PUBLIC SUMMARY.—The provider of a covered model shall—
(1) certify to the Administering Office that the model has passed a baseline security audit; and
(2) make available to the public a summary of the results of the audit, excluding any specific information that would materially assist a malicious actor in circumventing the model's safeguards.
(e) SCALING OF REQUIREMENTS.—The Administering Office shall establish audit standards that are proportionate to a covered model's capabilities, its manner of deployment, and the foreseeable risk it presents, so that the requirements of this section do not impose an undue burden on providers of lower-risk models; except that the protections described in paragraphs (2) through (5) of subsection (c) shall apply to every covered model and may not be reduced or waived on the basis of proportionality.
(f) RELATIONSHIP TO OTHER LAW.—The requirements of this section are in addition to any requirement applicable to a frontier model under this title and to any protection for minors afforded under Federal or State law, including State laws governing artificial intelligence chatbots.
TITLE III—WORKFORCE PROTECTION AND TRANSITION
SEC. 301. ADVANCE NOTICE OF ARTIFICIAL-INTELLIGENCE-RELATED EMPLOYMENT LOSS.
(a) EXTENDED NOTICE.—The Worker Adjustment and Retraining Notification Act of 1988 is amended to require not fewer than 90 days' advance notice of a covered plant closing or mass layoff, and to require notice to the Secretary of Labor and the Governor of each affected State.
(b) ARTIFICIAL INTELLIGENCE ATTRIBUTION.—A notice required under such Act shall state whether, and to what extent, automation or artificial intelligence systems are a cause of the employment loss.
SEC. 302. HUMAN REVIEW OF CRITICAL DECISIONS.
(a) IN GENERAL.—An artificial intelligence system may not serve as the sole basis for a critical decision.
(b) HUMAN AUTHORITY.—A qualified human reviewer shall have the authority and the practical ability to review, and to override, a determination in which an artificial intelligence system is used to make a critical decision.
SEC. 303. ARTIFICIAL INTELLIGENCE WORKFORCE TRANSITION FUND.
(a) ESTABLISHMENT.—There is established in the Treasury the Artificial Intelligence Workforce Transition Fund.
(b) FUNDING.—The Fund shall be financed by an assessment on covered developers and on large-scale artificial intelligence computing operations, in amounts established by the Administering Office in consultation with the Secretary of the Treasury.
(c) USE.—Amounts in the Fund shall be available for worker retraining, wage insurance, and transition assistance, with priority for—
(1) workers in occupations demonstrably contracting due to artificial intelligence; and
(2) rural and other communities hosting artificial intelligence computing infrastructure.
SEC. 304. PROCUREMENT PREFERENCE FOR HUMAN-COMPLEMENTARY SYSTEMS.
Where cost and performance are substantially comparable, a Federal agency shall prefer artificial intelligence systems that augment rather than replace the work of employees, and shall report annually on the employment effects of its artificial intelligence procurements.
SEC. 305. RIGHT TO NOTICE AND EXPLANATION.
A person subject to a critical decision in which an artificial intelligence system is used has the right to be informed of that use and to receive a plain-language explanation of the basis for the decision.
TITLE IV—MARKET INTEGRITY AND SYSTEMIC STABILITY
SEC. 401. SUBSTANTIATION OF CAPABILITY CLAIMS.
(a) SUBSTANTIATION REQUIRED.—A capability claim shall be substantiated by evidence at the time it is made and, where the claim concerns the capabilities of a frontier model, shall be supported by independent evaluation conducted under title II.
(b) UNLAWFUL PRACTICE.—A materially misleading capability claim, including a claim regarding revenue or cost savings attributable to artificial intelligence, is a violation of the securities laws, and the Securities and Exchange Commission may impose heightened penalties where such a claim was unsubstantiated when made.
SEC. 402. DISCLOSURE OF CIRCULAR FINANCING ARRANGEMENTS.
A covered developer, and any issuer whose valuation is materially dependent on artificial intelligence, shall disclose vendor-financing and circular-investment arrangements among artificial intelligence firms, including investments by an artificial intelligence firm in its customers or suppliers that may affect apparent demand for its products.
SEC. 403. SYSTEMIC EXPOSURE GUARDRAILS.
The Financial Stability Oversight Council shall—
(1) assess and publicly report the concentration of exposure of the financial system and of federally regulated retirement funds to a small number of firms dependent on artificial intelligence;
(2) conduct periodic stress tests of systemically important financial institutions against a sharp correction in artificial intelligence valuations; and
(3) establish prudential limits on—
(A) the extent to which systemically important financial institutions and federally regulated pension funds may hold assets whose value depends on artificial intelligence capabilities that have not been substantiated through independent evaluation; and
(B) the extent to which the construction of artificial intelligence computing infrastructure may be financed through leverage that presents a systemic risk.
SEC. 404. INVESTOR RISK DISCLOSURE.
Where the valuation of a public issuer is materially dependent on artificial intelligence capabilities that have not been substantiated through independent evaluation, the issuer shall include in its filings a standardized, plain-language disclosure of that dependence and the associated risk.
TITLE V—ARTIFICIAL INTELLIGENCE INFRASTRUCTURE; DATA CENTERS; AND PUBLIC ACCOUNTABILITY FOR PUBLIC INVESTMENT
SEC. 501. DEFINITIONS.
In this title:
(1) COVERED DATA CENTER.—The term "covered data center" means a data center facility used primarily to train or operate artificial intelligence systems that meets or exceeds a power-demand or capital-investment threshold established by the Administering Office.
(2) EXPANSION.—The term "expansion" means the establishment of a new covered data center or a material increase in the capacity of an existing covered data center.
(3) PUBLIC INVESTMENT.—The term "public investment" means any of the following, whether provided by the Federal Government or, where Federal support is conditioned on their treatment, by a State or political subdivision:
(A) A grant, loan, loan guarantee, tax credit, tax abatement, or tax exemption.
(B) Electric generation, transmission, or distribution infrastructure constructed at public expense principally to serve the facility.
(C) Water, wastewater, road, or other infrastructure constructed at public expense principally to serve the facility.
(D) The provision of public land or public resources at below-market value.
SEC. 502. PUBLIC OWNERSHIP INTEREST REQUIRED FOR TAXPAYER-FUNDED EXPANSION.
(a) IN GENERAL.—Public investment above a threshold established by the Administering Office may not be provided in support of the expansion of a covered data center unless the United States, or a public entity designated to hold the interest on behalf of taxpayers, receives, commensurate with the value of the public investment, one or more of the following:
(1) An equity or ownership interest in the facility or in the assets financed by the public investment.
(2) A royalty or revenue-sharing interest.
(3) Repayment of the public investment with a reasonable return.
(4) Another form of public return determined by the Administering Office to be of equivalent value.
(b) PROHIBITION ON UNCOMPENSATED SUBSIDY.—No public investment may be provided for the expansion of a privately owned covered data center that yields no ownership interest, return, or repayment to the public.
(c) PUBLICLY FUNDED INFRASTRUCTURE.—Electric generation, transmission, distribution, water, or other infrastructure constructed at public expense principally to serve a covered data center shall—
(1) remain publicly owned; or
(2) be paid for in full by the operator of the facility.
(d) DISPOSITION OF PUBLIC INTERESTS.—The Administering Office may hold, or designate an entity to hold and administer, ownership interests acquired under this section, including for the provision of public-interest computing capacity, and shall manage such interests for the benefit of taxpayers.
SEC. 503. DISCLOSURE OF PUBLIC SUPPORT.
Before public investment is approved for the expansion of a covered data center, the entity providing the investment shall make available to the public—
(1) an itemization of all Federal, State, and local public investment attributable to the facility;
(2) the projected effect of the facility on the rates of residential and small-business utility customers and on public water systems;
(3) the ownership structure of the facility; and
(4) the public ownership interest or return required under section 502.
SEC. 504. RATEPAYER, WATER, AND COMMUNITY PROTECTION AS CONDITIONS OF FEDERAL SUPPORT.
As a condition of receiving public investment, the operator of a covered data center shall—
(1) bear the full cost of the electric generation, transmission, and distribution capacity required to serve the facility, so that such costs are not shifted to residential or small-business ratepayers;
(2) publicly disclose the facility's projected direct and indirect water consumption and its cooling method, and obtain any required water-withdrawal permit before construction begins;
(3) enter into an enforceable community benefit agreement, including, where applicable, a commitment to monitor and protect local groundwater and private wells; and
(4) post a decommissioning bond sufficient to remove the facility and restore the site upon the cessation of its use.
SEC. 505. OVERSIGHT AND ACCOUNTABILITY.
(a) COMMITTEE CONSULTATION.—A decision by a Federal agency to provide public investment in support of the expansion of a covered data center is a covered governmental artificial intelligence decision for purposes of title I.
(b) AUDIT.—The Comptroller General of the United States shall audit public investments made under this title and the public ownership interests and returns received in exchange, and shall report the results to Congress and to the public not less frequently than annually.
(c) INFORMATION.—The Administering Office may require of a recipient of public investment such information as is necessary to verify compliance with this title.
SEC. 506. RECAPTURE.
If a covered data center that received public investment relocates, ceases operation, or fails to meet the employment or public-benefit commitments on which the investment was conditioned, the United States shall recapture the public investment, together with any agreed return, and any public ownership interest shall be enforced or adjusted accordingly.
TITLE VI—IMPLEMENTATION, ENFORCEMENT, AND FEDERALISM
SEC. 601. ADMINISTERING OFFICE.
(a) DESIGNATION.—The President shall designate or establish a Federal office to administer this Act, which shall coordinate with the National Institute of Standards and Technology on standards and evaluation, the Securities and Exchange Commission on title IV, the Department of Labor on title III, the Financial Stability Oversight Council on systemic risk, and the appropriate energy and infrastructure agencies on title V.
(b) COORDINATION WITH COMMITTEE.—The Administering Office shall provide the Committee such information and support as the Committee requires under title I.
SEC. 602. PENALTIES.
(a) CIVIL PENALTIES.—The Administering Office may impose civil penalties for violations of this Act, scaled to the annual revenue of the violator, with escalating penalties for repeat violations and for the suppression of safety-critical information.
(b) OTHER REMEDIES.—The remedies under this Act are in addition to, and not in lieu of, any other remedy available under law.
SEC. 603. RULE OF CONSTRUCTION; EFFECT ON STATE LAW.
Nothing in this Act shall be construed to preempt, or to authorize the preemption of, a law of a State or political subdivision that affords protections equal to or greater than those provided by this Act. This Act establishes minimum Federal requirements and does not limit the authority of a State to enact stronger protections.
SEC. 604. ADAPTIVE THRESHOLDS.
The Administering Office shall establish, and shall periodically revise, the compute and capability thresholds used to define a frontier model and a covered developer, so that such thresholds reflect the current state of artificial intelligence technology.
SEC. 605. AUTHORIZATION OF APPROPRIATIONS.
There are authorized to be appropriated such sums as may be necessary to carry out this Act, including sums necessary for the operation of the Committee and the Administering Office.