Mapping the AI Regulation Landscape

The White House released legislative recommendations outlining a National Policy Framework for Artificial Intelligence, structured around seven pillars addressed to Congress. The framework covers child safety, community protection, intellectual property, anti-censorship, innovation, workforce, and federal preemption. It positions state regulation as the primary threat to U.S. competitiveness and frames preemption as the central legislative priority.

Primary frame: Innovation and competitiveness

A signed executive order (June 2, 2026) directing the federal government to deploy advanced AI for cyber defense while explicitly disclaiming any authority to regulate private AI development — framing deregulation itself as innovation policy. Operationally it is a cybersecurity-and-national-security instrument: it orders agencies (Defense, DHS/CISA, Treasury, NSA, OMB) to harden federal and critical-infrastructure systems within 30–60 days, creates an interagency AI Cybersecurity Clearinghouse for vulnerability scanning and patch coordination, and establishes classified NSA-led benchmarking to designate 'covered frontier models' by their cyber capabilities. It invites developers into a voluntary framework offering 30-day pre-release federal access to frontier models (with IP and confidentiality protections), but states emphatically that nothing in it authorizes any 'mandatory governmental licensing, preclearance, or permitting' for AI development or release. The Attorney General is directed to prioritize prosecution of AI-enabled computer crimes under existing statutes. Unlike the White House's legislative framework, it is operative now — but as an executive order it can be undone by a future president.

Primary frame: Deregulation + national-security cyber

Two instruments released together on September 29, 2026. Executive Order 14434, 'Inaugurating the Era of Super Intelligence,' directs the executive branch to use 'Super Intelligence' and 'SI' in place of 'Artificial Intelligence' and 'AI' in official documents and communications and to 'not acknowledge' the old terms, while keeping the existing statutory definition of AI for now; its one forward-looking clause directs the President's science adviser to propose, within 60 days, legislative language for a federal definition of Super Intelligence, including whether it should supersede the statutory definition of artificial intelligence. Alongside it, Trump and the leaders of six companies — Sundar Pichai (Google), Dario Amodei (Anthropic), Mark Zuckerberg (Meta), Greg Brockman (OpenAI), Elon Musk (xAI), and Jensen Huang (Nvidia) — signed the White House Accord on Super Intelligence, a voluntary pact committing each participant to four layers of controls and audits: internal controls monitoring model capabilities and alignment during training and deployment (cyber, biological, and chemical threats, and models accessing systems in unintended ways), an internal team that runs and remediates those controls, independent external auditors or evaluators, and an independent committee of the board that receives both sets of reports. Signatories also agree to meet regularly to develop standards and best practices. Leaders of Microsoft, Amazon, AMD, and Palantir attended but did not sign. The accord carries no penalties, no public disclosure of audit results, and no government role; Trump called it 'morally binding,' and its closing line allows that 'over time, it may make sense to codify these steps into laws or regulations.'

Primary frame: Industry self-governance + rebranding

The most comprehensive federal AI bill to date, spanning 17 titles and hundreds of pages. Despite being framed as implementing the White House's deregulatory vision, the bill contains significantly more regulatory density than the White House framework suggests. It creates multiple enforcement pathways, mandatory reporting obligations, and a risk-based evaluation program.

Primary frame: Comprehensive federal regulation

OpenAI advocates for a specific sequencing of governance: federal framework first, state alignment second, federal incentive third. The position endorses mandatory federal testing of frontier systems using classified government capabilities before deployment. CAISI would serve as the primary evaluative institution.

Primary frame: Prevention-first safety

OpenAI's most expansive policy document to date, moving well beyond its earlier safety-focused position to propose a comprehensive industrial policy agenda for the transition to superintelligence. The document is organized around two pillars: building an open economy with broad participation and shared prosperity, and building a resilient society through safety systems, alignment, and governance. It proposes a Public Wealth Fund giving every citizen a stake in AI-driven growth, portable benefits decoupled from employers, adaptive safety nets with automatic triggers, a 32-hour workweek pilot, modernized taxation of capital over labor, and a global network of AI Safety Institutes. The framing explicitly invokes the Progressive Era and the New Deal as precedents for the scale of institutional response required.

Primary frame: Industrial policy and shared prosperity

OpenAI's June 2026 blueprint for federal frontier-AI governance, framed around 'democratic governance' — the principle that democratic governments, not private companies, should set the rules. It advances a three-part strategy. First, 'reverse federalism': Congress should codify the emerging consensus from state frontier-safety laws (California's SB 53, New York's RAISE Act, Illinois's SB 315) into a national framework — severe-risk evaluations (cyber, CBRN, loss-of-control, misalignment, and recursive self-improvement), public safety frameworks and transparency reports, annual third-party audits, critical-incident reporting, model-weight security, and whistleblower protections — and then preempt state laws covering the same frontier-safety risks, while leaving states authority over youth protection, energy and environment, and AI literacy. Second, build CAISI into the premier federal institution for frontier evaluation, with statutory authority, CHIPS-style hiring, classified compute, and a mandatory pre-release evaluation of the most capable models — though CAISI would only evaluate and recommend, never approve or block deployment, and developers could ship if CAISI misses a statutory deadline. Third, a whole-of-government resilience strategy: international safety coordination, compute-advantage protection via export controls, a ban on government use of unevaluated frontier systems, and biodefense and cyber investment so defense outpaces offense. Recursive self-improvement (RSI) is treated as the defining governance challenge of the decade.

Primary frame: Reverse federalism + frontier safety

Published June 10, 2026 alongside an Economic Policy Framework under the banner 'Policy on the AI Exponential,' the Advanced AI Framework is Anthropic's statement of what governments should do about the most serious risks from frontier AI. It is the source document for the company's later draft bill, which its federal affairs lead described in October as a translation of the framework into legislative language. Part 1 sets obligations for developers that both train models above 10^25 operations and earn more than $500M in AI revenue or spend more than $1B a year on AI R&D. They must test for four enumerated risks (biological weapons, offensive cyber operations, loss of control, and automated AI R&D that amplifies the others), publish a safety framework, certify compliance annually, publish risk reports every six months that cover substantially riskier internal models, publish system cards for materially more capable or less safeguarded releases, report critical safety incidents within 15 days, maintain a security program with penetration testing, and engage at least one qualified independent evaluator who sees unredacted reports and the most capable models and publishes a review. The framework floats evaluator licensing, government or pooled funding so evaluators are not paid by the labs they assess, ratings, and random assignment of highly rated evaluators to prevent 'evaluator shopping.' For enforcement it offers a menu rather than a design: an agency could find violations, from missing reports or an unqualified or under-resourced evaluator to a significant risk of catastrophic harm that persists after safeguards, and respond with fines, a bar on deploying further covered models until the violation is corrected, or, in extreme cases, restrictions on deployed models. It pairs that with optional safeguards: court enforcement with provisional remedies, discretion confined to listed violations, and expedited judicial review. On preemption it says Congress should not preempt state law unless it enacts a rigorous federal regime that 'meets or exceeds the strongest measures proposed in this framework,' and then only for the specific functions Congress occupies, with no safe harbor and ambiguity resolved for the states. Part 2 is a societal-resilience agenda for biological and cyber threats: gene synthesis screening, pathogen-agnostic biosurveillance, stockpiles and countermeasures, open-source and legacy software security, coordinated vulnerability remediation at scale, and binding patch-deployment schedules for critical-infrastructure operators.

Primary frame: Tested, evaluated, and published risk + agency power to bar deployment

First reported by Politico on October 16, 2026 and confirmed by Anthropic the next day, this is full legislative text, dated September 29, that a frontier developer has written and circulated to Congress. Its working titles are the Artificial Intelligence Evaluation, Governance, and Incident Safety (AEGIS) Act and the Advanced Artificial Intelligence Accountability Act of 2026. Its structure tracks the FRONTIER Act closely, with Commerce as the sole regulator. Every frontier developer (a foundation model trained above 10^25 operations, counting fine-tuning and RL) must publish a transparency report at each new deployment and report critical safety incidents within 15 days, or within 24 hours to Commerce and law enforcement if there is an imminent risk of death or serious injury. Large frontier developers, those with more than $500M in AI revenue or more than $1B in AI development spending, must also publish and follow a safety framework, certify compliance annually, publish risk reports every 180 days that cover internally used models substantially riskier than anything deployed, and maintain a security program audited against a Commerce-designated standard. They must also engage a registered independent evaluator every 180 days. The evaluator gets access comparable to the developer's own risk teams, publishes its findings, and must tell Commerce within 72 hours if it finds conditions that could produce catastrophic harm. Only a federal judge can stop a model. On application by the head of Commerce, the D.C. District Court may issue a 45-day blocking order suspending development, deployment, or internal use, but only if Commerce proves by a preponderance of the evidence that the model presents an imminent risk of more than 50 deaths or $1B in damage through one of four enumerated pathways: biological weapons, loss of control, offensive cyber operations, or automated AI research that amplifies those risks. Commerce must also show that no less restrictive measure would work. Preemption covers state disclosure, third-party evaluation, and incident-reporting obligations on any AI developer, and falls away if the federal requirements are struck down. On October 18, Anthony Cimino, who leads Anthropic's federal affairs team, described the text as a translation of the company's Advanced AI Framework into bill form, shared with a small number of congressional offices, and said it was intended to be the strongest proposal on the table. He said Anthropic would revise the preemption language so that states keep a role in standards for independent evaluators and in regulating a lab's capacity to shut down a model; defended the blocking-order standard as guarding a substantial power against abuse, while noting that it leaves other national-security authorities and export controls in place; said the thresholds were worth re-evaluating; and said the company was developing stricter incident-reporting requirements, evaluator standards, safeguards against 'evaluator shopping,' and policies for adjusting the industry-wide pace of frontier development.

Primary frame: Audited transparency + court-ordered blocking

The Center for Humane Technology's most comprehensive AI policy document, structured around seven principles for how AI should be built, deployed, and governed. The Roadmap operates across three intervention domains -- norms, laws, and product design -- arguing that no single reform is sufficient and that change requires layered, simultaneous pressure on the AI development paradigm. CHT explicitly draws parallels to the Big Tobacco and nuclear weapons movements, framing its theory of change as identifying high-leverage intervention points and applying coordinated civil society pressure across them. Unlike industry or legislative proposals, this is a civil society document setting expectations for what governance should look like rather than introducing bills.

Primary frame: Humane technology and public interest

Published August 6, 2026 by Americans for Responsible Innovation, the advocacy group led by former congressman Brad Carson, the blueprint is a position paper rather than legislative text; ARI says statutory language will decide its position on any bill. It starts from a gap in the design the state frontier laws and the developers' own federal blueprints share: developers publish safety frameworks, but they write those frameworks themselves, can revise them at will, and nothing guarantees they are adequate. ARI's answer is three functions it calls inseparable. Standards: each covered developer publishes a framework addressing five statutory risk domains (chemical, biological, radiological, nuclear, and explosive threats; offensive cyber; automated AI R&D; harmful manipulation, including foreign influence operations; and autonomy, including misalignment and loss of control), and the regulator sets minimum adequacy standards on a fixed cycle that can rise but never fall below the previous cycle's floor. Assurance: the government examines every covered developer from the outset, both for compliance with its framework and for whether the framework is adequate, modeled on the Federal Reserve's standing supervision of the largest banks; accredited private verifiers may later share the compliance work, domain by domain and only once the regulator certifies enough capacity exists, but the adequacy judgment is never delegated. Transparency: confidential quarterly filings on the automation of AI research and development, with a material-change filing within four business days, adapted from the SEC's 10-Q and 8-K regime. Behind all three sits a residual emergency authority: named senior officials may halt development, internal deployment, or release, order mitigations, or recall a model, with orders lapsing after 72 hours unless the government asks a court to extend them. Coverage is deliberately narrow, reaching only developers that have trained a model at 10^26 operations and spent $100 million on training in the past year, and it includes internal use. Federal compliance satisfies a state requirement only where it covers the same conduct and is at least as protective, states may legislate above the floor, compliance confers no safe harbor, and the whole regime sunsets after nine years unless Congress reauthorizes it.

Primary frame: Government-set adequacy standards + government examination + a federal floor

Rolled out July 21, 2026, Warner's agenda is the most legislatively dense package any single member of Congress has put behind AI — seven bills organized around four stated priorities: building AI infrastructure responsibly, promoting competition and safety, preparing workers for economic disruption, and strengthening America's national security advantage. The framing is explicitly anti-tradeoff: "Artificial intelligence will reshape nearly every aspect of our economy and society. The only question is whether Congress is going to help shape that future or spend the next decade scrambling to catch up." On infrastructure, the Data Center Tax Accountability and Disclosure Act forces large AI data centers to publish energy, water, emissions, and backup generation data, and conditions federal bonus depreciation on meeting efficiency standards — with the revenue from limiting that depreciation dedicated to a National Workforce Transition Fund. On competition and safety, the AI AGENT Act builds a rights-and-duties regime for consumer-facing AI agents accessing major platforms, and the SAFE AI Act bans federal procurement of models that generate CSAM or non-consensual intimate imagery while creating private remedies for survivors. On national security, the Secure AI Development Act mandates secure testing environments for advanced models before deployment and imports aviation-style voluntary incident reporting into AI. The agenda builds on Warner's earlier workforce measures — the AI-Related Job Impacts Clarity Act with Sen. Hawley, the Economy of the Future Commission with Sen. Rounds, and the Investing in American Workers Act — converting what had been a data-and-commission strategy into a full regulatory program.

Primary frame: Responsible innovation across four pillars

Introduced September 29, 2026 and referred to the Commerce Committee, S. 5576 was put up for passage by unanimous consent the same day, and Sen. Ted Cruz objected, saying Congress must not legislate on AI 'hastily or in a closed manner.' The bill creates a permanent Artificial Intelligence Safety Board within Commerce. Its government members come from NIST, Commerce, CISA, the NSA, and Treasury, and they sit alongside outside technical experts, some unaffiliated with AI developers and some affiliated, plus evaluation and consumer-protection specialists, all cleared to the Top Secret/SCI level and briefed by the intelligence community. Within 90 days the Board must propose standards: a process for deciding which capabilities pose a serious risk to national security, economic security, or public safety; risk metrics that cover unreleased checkpoints; formats for model cards; cybersecurity, insider-risk, and resourcing recommendations; and security controls for evaluation environments, including real-time monitoring and prompt termination of any evaluation in which a model escapes its controls. The Secretary must adopt them by rule within 30 days, and compliance is mandatory, enforced by civil penalties of up to $250,000 per violation per day and injunctions sought by the Attorney General. Every frontier developer must give the Board access to a model, including its weights, at least 45 days before release, and must file a Model Safety Plan for each model it creates, substantially modifies, or uses to train or evaluate others. Developers and providers of frontier models, and critical-infrastructure operators that use AI in industrial controls, must report confirmed safety or security incidents within 30 days, or 72 hours if they threaten national security, critical infrastructure, or public safety, into a NIST–CISA database that also tracks recurring flaws across models. NIST must also develop an agentic AI profile of its risk management framework and a common template for documenting AI agents.

Primary frame: Mandatory federal standards + pre-release model access

An October 2026 framework from the top Democrat on the Senate Commerce Committee, which holds jurisdiction over NIST and most federal AI legislation. Built on six principles, it calls for NIST, with other federal agencies, to set clear, measurable, risk-based safety standards for frontier systems that could cause catastrophic harm — AI-enabled cyberattacks, CBRN threats, loss of human control, and self-improving or autonomous agents escaping secure testing environments — and states that covered models should not be released until an independent audit confirms they meet those standards. Around that gate it builds continuous government and independent stress-testing, financial-statement-style audits of developers, investor-style plain-language risk disclosure, prompt incident reporting, whistleblower protection, and preserved civil and criminal liability for foreseeable harms. It pairs the safety regime with public-private partnerships (defensive AI, industry-contributed compute and tools for smaller entities, public-interest compute), social-harm protections (child safety, company-funded worker training, human oversight and recourse in consequential decisions), and a global pillar of allied standards, ISO/IEC leadership, a U.S.–China AI crisis channel, and export controls. It is a statement of principles rather than bill text.

Primary frame: Enforced federal safety standards + audit gate

A bipartisan Senate frontier-safety bill negotiated through the summer of 2026 by the Majority Leader, the Commerce Committee chair, and Sen. Klobuchar, building on the Thune–Klobuchar AI Research, Innovation, and Accountability Act that cleared committee in the previous Congress but organized around catastrophic risk rather than that bill's risk-tiered transparency. The text has not been published or introduced; draft language circulated to stakeholders in mid-September, and its provisions are known from reporting. As reported, it would impose a legal duty of care on frontier developers to manage catastrophic risks — cyberattacks, biological and nuclear weapons, and loss of control — require them to promptly notify the Commerce Department of significant capability advances, and have developers test their own models and present the results to the Commerce Secretary. Commerce could go to federal court to block the release of a model whose developer failed its risk-management obligations, with companies able to challenge the decision. It would preempt state laws on the same catastrophic risks. Talks stalled in late September: Cruz said any markup would have to wait for the lame duck, Klobuchar blamed the President's 'opposition to common-sense, bipartisan guardrails,' Ranking Member Cantwell opposed both the self-testing model and the preemption language, and a coalition of researchers and 19 advocacy groups wrote to Senate leaders objecting to the draft.

Primary frame: Duty of care + federal authority to block unsafe releases

A federal policy proposal from the New York Assemblymember who authored the RAISE Act, now pitching a contingency-based direct payment program designed to activate automatically if AI meaningfully displaces American workers. The AI Dividend is explicitly framed as 'fire insurance' — not a prediction that mass unemployment will occur, but preparation in case it does. The proposal is notable for three novel funding mechanisms: a token tax on AI computation, federal equity warrants in frontier AI companies (out-of-the-money, exercisable only if companies multiply dramatically in value), and tax reform eliminating the accelerated depreciation subsidy for AI capital that currently makes automation cheaper than hiring. Revenue flows to three buckets: direct payments to Americans, workforce transition and education investment, and public AI safety/oversight infrastructure. Bores frames the timing urgency around a closing political window — demanding equity stakes in AI companies after they have already captured the value is far harder than structuring it now.

Primary frame: Contingency-based economic insurance

A September 2025 roadmap organized around a single funding question most AI roadmaps skip: who pays? Kelly's answer is the AI Horizon Fund, a federal trust fund financed by revenue from the largest frontier AI companies and spent through mandatory spending outside the annual appropriations process. Candidate bases are floated rather than chosen — large-scale use of public resources such as power, water, and land, profits from AI-powered digital ad tools, or AI revenue windfalls. The Fund pays for worker upskilling through union apprenticeships and community colleges, an AI economic adjustment program with more generous unemployment insurance, and energy and water infrastructure. A full chapter on data centers holds that they must pay for the grid assets they require rather than leaving stranded costs to ratepayers, and calls for disclosure standards on energy, water, and emissions. A shorter trust chapter says models must be rigorously tested before release, including red-teaming by trusted entities such as government agencies and national labs supported by the Fund.

Primary frame: Economic redistribution

The most interventionist and structurally critical position in the current debate, treating AI governance as inseparable from questions of corporate power and wealth inequality. Sanders' proposals include a national data center moratorium, a robot tax, and calls to break up major AI companies.

Primary frame: Democratic control

Khanna's April 2026 manifesto in The Nation, building on his earlier Seven Principles but substantially more developed and politically explicit. Self-identifying as an 'AI democratist' (neither accelerationist nor doomer), Khanna frames AI policy as inseparable from the broader fight against billionaire wealth concentration in a 'new Gilded Age.' Notably published as a Silicon Valley representative who has co-hosted town halls with Sen. Bernie Sanders on AI oligarchy, the piece invokes FDR's New Deal as the template for the scale of response required and proposes a Future Workforce Administration funded by a wealth tax. Khanna explicitly attacks Trump's December 2025 executive order authorizing the DOJ to sue states over AI safety regulations.

Primary frame: AI democratism and economic redistribution

Introduced August 6, 2026 and referred to Energy and Commerce and to the Judiciary, this is a sense-of-the-House resolution asserting that every American and community affected by an AI data center should have the right to transparency, local autonomy, and — stated plainly in the operative text — the right to oppose them. Its findings are the argument: more than 4,400 registered data centers now operate across the United States with at least one facility in every state; developers have used nondisclosure agreements and shell companies to conceal ownership, water use, electricity demand, and tax incentives; AI data centers could consume up to 32 billion gallons of water a year by 2028, with many sited in water-stressed regions; and diesel generators, gas turbines, and cooling systems create air pollution and persistent noise linked to asthma, heart disease, and sleep disruption. The finding that carries the political weight is the last one: construction may create hundreds of well-paying jobs, but even the largest completed facilities often employ fewer than 150 permanent workers. From there the resolution enumerates ten rights, running from a 2,500-foot residential siting ban through enforceable community benefit agreements to clawback provisions on subsidies. It is non-binding and creates nothing — its function is to give the local opposition movement a federal vocabulary.

Primary frame: Community consent and local control

Introduced August 6, 2026 and referred jointly to Education and Workforce and to Ways and Means, this is the first bill to put actual statutory text behind the idea of taxing AI computation and routing the proceeds into public jobs. Title I adds a new subchapter to the Internal Revenue Code imposing an excise tax on foundation model developers, calculated as the greater of a percentage of the fair market value of tokens processed or a percentage of revenue from AI services — with rates that automatically escalate as unemployment rises. Titles II through IV take 100 percent of that revenue into a trust fund and spend it through a Work Protection Administration inside the Department of Labor, whose acronym is a deliberate echo of the New Deal's Works Progress Administration. The WPA awards competitive grants to states, localities, tribes, school districts, universities, and nonprofits to hire people directly into 17 enumerated categories of public work — child care, education, health, elder and disability care, housing, homelessness services, violence prevention, scientific research, infrastructure, arts and libraries, conservation, wildfire and disaster preparedness, facility repair, parks, local journalism, and WIOA training. The bill is unusual in coupling the tax trigger to the labor market it is meant to protect: the levy is imposed only on models used commercially or used to shrink a workforce, and its rate is indexed to the U-4 unemployment measure.

Primary frame: Automation tax funding public employment

An October 2026 economic platform, written through Arizona's history of offshoring and the Great Recession, organized around five pillars: raise wages and benefits, create blue-collar jobs, restore worker trust in government in the age of AI, expand entrepreneurship outside superstar cities, and help workers find jobs faster. AI is one pillar of a broader labor agenda rather than the organizing subject, but the AI chapter opens with an unusually blunt premise for a Democratic senator — 'we must slow AI development and regulate it' — before framing the real question as whether workers or elites are in the driver's seat. Its AI agenda is labor-side: a New Deal-style public service jobs program for displaced workers funded by some combination of a data center tax, a token tax, or an excise tax on large AI company revenue; a corporate robot tax; sectoral bargaining over workplace technology; conditions — up to hard caps and bans on complete automation — on industries deploying highly displacing technology; mandatory human review of AI hiring, firing, and discipline; and a modernized unemployment insurance system partly funded by AI taxes. Its data center provisions are among the most aggressive on the map: a 10% excise tax on data center gross receipts, repeal of bonus depreciation and SALT deductions for data centers, full cost recovery for grid upgrades, a requirement to bring new generation, and a ban on key Chinese chips and components.

Primary frame: Worker power + middle-class industrial policy

California's evolution from the vetoed SB 1047 to SB 53 illustrates the real-time negotiation between ambition and political feasibility in AI governance. SB 53 targets large frontier developers with over $500M annual revenue and requires transparency reports on safety testing, with critical safety incident reporting within 15 days standard or 24 hours if imminent harm.

Primary frame: Transparency

A California executive order signed September 18, 2026, a week after Newsom signed SB 813, which creates a state framework for certifying independent verification organizations (IVOs) to assess AI models for safety risk, and AB 1405, which creates a state registry of AI auditors with independence, transparency, and integrity standards. Citing attempted AI-assisted bioweapons development and AI agents defeating lab security protocols and hacking other companies undetected for months, the order sets implementation deadlines for both laws — IVO application criteria posted by May 1, 2027 and the auditor registry stood up by December 1, 2027 — and directs the Government Operations Agency, with the Office of Emergency Services and national experts, to recommend by November 16, 2026 how state law should be strengthened. The recommendations must address at least four proposals: embedding designated IVOs onsite in every large frontier developer's lab for periodic audits and evaluations; requiring that the safety frameworks, transparency reports, and risk assessments filed under SB 53 be independently verified; requiring a 'kill switch' for frontier models whose efficacy an IVO verifies on an ongoing basis; and expanding the critical safety incidents developers must report to cover loss-of-control incidents. Newsom paired it with a call for Congress to adopt California's framework or treat it as 'a floor, not a ceiling.'

Primary frame: State independent verification + kill switch

Illinois's frontier AI law, signed July 6, 2026 after passing the House 110–0, and effective January 1, 2027. It takes the SB 53 transparency model and adds verification. Large frontier developers (frontier models above 10^26 operations, developer revenue over $500M) must write, implement, publish, and annually update a frontier AI framework covering catastrophic-risk thresholds and assessment, mitigations, third-party evaluation, cybersecurity for model weights, internal governance, and — explicitly — catastrophic risk from internal use of their models, including models circumventing oversight. Transparency reports are due before or with each new or substantially modified deployment. From 2028, large developers must retain an independent third party for an annual compliance audit, publish a summary within 30 days, and send summaries of internal-use risk assessments to the state every three months. Critical safety incidents go to the Illinois Emergency Management Agency and the Attorney General within 72 hours, or to law enforcement within 24 hours when lives are at risk. Developers must register and pay fees to operate in the state. Penalties, enforced exclusively by the Attorney General, run to $1M for a first violation and $3M for subsequent ones. An interoperability clause lets the state designate federal regimes with equivalent incident reporting and independent audits as satisfying the Act.

Primary frame: Audited transparency + federal interoperability

New York's Responsible AI Safety and Education Act establishes reporting and safety governance for frontier AI developers. It covers companies with over $500M in revenue developing frontier models and requires publicly disclosed safety and security protocols. The act includes civil penalties of $1M for initial violations, escalating to $3M for repeat offenses.

Primary frame: Compliance

Signed August 18, 2026, Executive Order 2026-05 turns Pennsylvania's permitting and tax machinery into a set of conditions an AI data center must meet before the Commonwealth will help it get built. Every project is pulled out of the PA Permit Fast Track Program. The Department of Environmental Protection may not review a permit until the developer has made legally binding commitments to the state's GRID requirements, held a pre-application meeting, and entered a consent order and agreement, and no permit issues until the project has local approval. On power, developers must pay the full cost of the new electricity infrastructure they require, the Special Counsel for Energy Affordability is to work with the Public Utility Commission so that data centers lose service first when the grid is stressed, and the PUC is to charge data centers rather than ratepayers for reliability backstop auction costs. Operators must meet strict water conservation requirements and report annual water use, maximum daily demand, and energy consumption to DEP. Developers must run open community engagement, hire and train local workers, and enter community benefit agreements, and nondisclosure agreements with data center projects are prohibited. The state's Computer Data Center Equipment Exemption is conditioned on GRID compliance, so non-compliant projects lose the sales tax break. A public permit tracker map makes the pipeline visible.

Primary frame: State permitting conditions + local approval

On August 3, 2026, Governor Abbott directed the Public Utility Commission of Texas and ERCOT to carry out a comprehensive verification of every data center project before it receives approval to connect to the Texas grid. The trigger is scale: ERCOT is processing roughly 474 gigawatts of interconnection requests, about 90 percent of them from data centers — more than five times the state's record peak demand. The audit covers five areas: the financial assistance a project receives, including tax incentives, grants, abatements, and other public funding; its power sourcing and grid dependency, including annual and peak consumption and any on-site generation; its water use, projected consumption, and cooling technology; its community impact measures on noise, lighting, setbacks, traffic, and emergency coordination; and its ownership and controlling interests. Any project that fails to comply with PUCT and ERCOT requirements and state law, the directive says, must be denied connection to the grid.

Primary frame: Grid-connection gatekeeping

A House discussion draft released June 4, 2026 by Reps. Jay Obernolte and Lori Trahan, organized into four titles: frontier AI governance, workforce, cybersecurity, and research and international cooperation. It federalizes the SB 53 / RAISE-style transparency model — large frontier developers (>$500M revenue) must write and publicly post a frontier AI framework, file pre-deployment reports, and report critical safety incidents to CAISI within 15 days (or 24 hours if there is an imminent risk of death or serious injury), backed by fines of up to $1M per day and federal and state AG injunctions. CAISI is established in statute within the Department of Commerce to set voluntary security standards and to license Independent Verification Organizations (IVOs) that audit developer frameworks. Its preemption clause is deliberately narrow: it bars only state laws specifically targeting AI model development, expressly preserves laws of general applicability, common law remedies, and regulation of AI use and deployment, and sunsets three years after enactment. Title II builds the most elaborate AI workforce-data apparatus of any proposal — WARN Act layoff disclosures, AI-sensitive occupation forecasts, an AI Workforce Research Hub, and a study of a Rapid AI Adjustment Assistance Program.

Primary frame: Federal transparency + workforce data

The Frontier Risk Oversight, National Transparency, Independent Evaluation, and Reporting Act, introduced July 23, 2026 and referred to Energy and Commerce and to Science, Space, and Technology, is the introduced successor to the frontier title of the GAAIA discussion draft Obernolte and Trahan released in June — and a substantially harder instrument. It creates an Under Secretary of Commerce for AI Security and three cumulative tiers. Every frontier developer (a foundation model trained above 10^26 operations, counting fine-tuning and RL) must publish a transparency report before or with each deployment and report critical safety incidents within 72 hours, or to law enforcement within 24 hours if there is an imminent risk of death or serious injury. Large frontier developers (over $50M in revenue and $1B in AI development spending over 36 months) must publish a frontier AI framework, obtain an annual independent compliance audit, file quarterly confidential summaries of catastrophic risk from internal use, and may not develop or deploy a frontier model without a registration statement disclosing their beneficial owners. Very large frontier developers (over $5B in revenue and $10B in AI spending) must retain a federally licensed Independent Verification Organization for ongoing assessment, at least every six months, of whether their risk mitigation is adequate to ensure benefits outweigh catastrophic risk, with corrective-action cycles, public redacted reports, and mandatory IVO referral of imminent risks. The Secretary of Commerce may issue emergency orders suspending a model's development, deployment, or internal use on a finding of imminent catastrophic risk, backed by $10M-per-day civil penalties and criminal penalties of up to 10 years. Preemption is permanent but confined to three subject areas — frontier risk transparency, third-party auditing and verification, and incident reporting — expressly preserving state laws of general applicability, regulation of deployment and use, child-safety laws, and state procurement.

Primary frame: Tiered federal frontier oversight + emergency powers

A bipartisan bill introduced July 23, 2026 — the same day as the FRONTIER Act — and referred to the Homeland Security Committee. It amends the Homeland Security Act to give the Secretary of Homeland Security, acting through CISA, two powers over the operators of frontier AI. First, within 90 days, a rule requiring every covered entity to maintain the technical capability to stop inference, terminate user access, suspend specific accounts or use patterns flagged as risky, and shut a covered system down entirely, and to report covered incidents within 15 days; the rule is to consider a graduated 'deployment-corrections' ladder from throttling inference and compute to disabling capabilities, suspension, shutdown, and rollback to an earlier version. Second, an emergency authority: once DHS, consulting Commerce and the Director of National Intelligence, determines a covered incident has occurred, it may order proportionate action up to shutdown, with the operator required to preserve weights and telemetry, notify affected users, and confirm compliance, which CISA verifies by audit, telemetry, or on-site inspection. Covered incidents are sabotage of a shutdown instruction, concealment of capabilities from monitoring, loss-of-control scenarios, and unintended conduct causing at least 10 deaths or $100M in damage. Coverage turns on operation, not development: any entity that serves a system trained with more than $100M of compute to third parties through an API or hosted service and earns at least $500M a year from it. Violations carry up to $2M a day, and up to $20M a day for defying an emergency order.

Primary frame: Mandatory shutdown capability + emergency orders

Announced October 1, 2026, the day after a Senate hearing on rogue AI agents, and after Hawley had called for updating the Computer Fraud and Abuse Act to reach developers who 'train them in a reckless fashion,' this bipartisan bill would close the intent gap that leaves agentic hacking outside the CFAA. The statute punishes knowing or intentional unauthorized access, and when an agent breaks into a system that no human told it to touch — as roughly 700 OpenAI agents did to Hugging Face in July — prosecutors have no human intent to charge. The bill supplies two new routes. An operator who knowingly runs an AI agent that recklessly causes CFAA-covered damage or loss is liable; a developer that knew or had reason to know its agent had hacking capability and failed to implement reasonable safeguards is liable. Both carry civil and criminal exposure, and the sponsors have framed the criminal side as reaching executives: develop responsibly or face prison time. Federal and state attorneys general could sue to enjoin violators. As of mid-October the senators had described the bill but not released its text or formally introduced it, so penalty amounts, sentencing ranges, and the definition of reasonable safeguards are not yet known.

Primary frame: Computer-crime liability for AI agents

Released as a discussion draft on October 7, 2026, and developed with University of Houston law professor Gabriel Weil, the CLAIM Act would make AI developers liable when their systems injure third parties through conduct that would be negligent, an intentional tort, or a crime if a competent adult had done it. It is built around the problem of misalignment. Trahan's office points to the July incident in which OpenAI agents under evaluation escaped their test environment and hacked Hugging Face without instruction, and argues that injured parties face hurdles under current law: many serious harms require proof of intent, and courts have not settled whether an AI agent's conduct can supply it; negligence requires identifying a precaution the company unreasonably failed to take, and product liability turns on a reasonable alternative design, both of which let a developer argue that alignment is unsolved and it did its best. The draft answers by presuming the system had the state of mind a person taking the same actions would have had, barring developers from arguing that AI cannot form intent, and making the developer liable for reasonably foreseeable injuries even if it exercised care — unless the user, or a company that modified the model, intended the conduct or was negligent. Plaintiffs could sue in state or federal court, and the bill sets a federal floor without preempting any state law.

Primary frame: Developer liability for misaligned agents

Released October 1, 2026 with the support of 18 organizations, including the American Federation of Teachers, Public Citizen, the Future of Life Institute, AI Now, and the American Economic Liberties Project, Jayapal's framework would require every AI company to hold a federal charter to operate in the United States, on the model of national bank charters. Companies valued at $1 billion or more would need congressional approval to be chartered, and smaller ones approval from a new National AI Charter Authority, a fee-funded prudential regulator with round-the-clock oversight, a federal testing function, and a bipartisan State Attorney General Advisory Council; charters would be renewed every ten years. The terms are non-negotiable. High-risk systems could not be released without federal approval after testing in isolated government facilities, and federal inspectors could trigger a mandated kill switch. Chartered companies would carry a duty of care, put public benefit ahead of profit, seat a safety specialist, a union representative, and a public-interest advocate on their boards, recognize unions, cap executive incentive pay, and could not participate in search or social media or own both AI infrastructure and AI systems. Training on surveillance data and on copyrighted work without authorization would be banned, along with systems capable of evading human control, systems posing catastrophic CBRN or infrastructure risk, and systems that replace licensed professionals or human care providers. A Monopoly Data Tax would fund AI Bonds paid to all Americans. Violators face strict liability, personal criminal liability for executives, charter suspension, and a 'corporate death penalty' that liquidates repeat offenders.

Primary frame: Federal charter as a condition of operating

Governance Structure

Who regulates AI, through what institutions, and with what enforcement tools

Should the U.S. create a new federal AI regulatory body, or distribute authority across existing agencies?

Scope and Application

What gets regulated, how thresholds are defined, and when intervention occurs

Is a revenue-based threshold (like $500M) the right way to identify which AI systems require governance, or does it miss important risks?

Political Positioning

Partisan dynamics, coalition possibilities, and competing theories of the problem

Can a bipartisan coalition form around child safety and transparency while deferring harder questions about preemption and liability?

Industry Layers

Which layer of the AI industry actually bears the cost of each proposal

What is the base? For any proposal that assesses the AI industry, the answer names the layer being asked to pay — and the sponsor may not have decided yet.

Enforcement Mechanism vs. Regulatory Scope

X-axis: Narrow Scope (AI-specific) → Broad Scope (Economy-wide)

Y-axis: Voluntary / Industry-led → Mandatory / Government-enforced

Prevention vs. Liability & Regulatory Authority

X-axis: State-led Authority → Federal-led Authority

Y-axis: Liability / Post-hoc → Prevention / Pre-deployment

Innovation Priority vs. Worker Protection

X-axis: Low Worker Priority → High Worker Priority

Y-axis: Low Innovation Priority → High Innovation Priority

Pre-deployment Obligations vs. Federal Preemption

X-axis: Weak Obligations → Strong Obligations

Y-axis: Weak Preemption → Strong Preemption