The AI Industry

Great American AI Act Draft: What Three-Year State Preemption Would Change

A bipartisan House draft would set federal rules for frontier AI developers and pause state laws on model development for three years.

Tobias Reyes

AI Industry & Policy Analyst

Published 7 min read
A breathtaking view of the US Capitol at sunrise in Washington, DC.
In this story 9 sections

The Great American AI Act is a bipartisan House discussion draft, released June 4, 2026, that would set federal rules for large frontier AI developers. It would also pause state laws on how AI models are developed for three years. It has not been filed as a bill, and critics say the preemption clause goes too far.

This explainer is for AI company leaders, compliance teams, and readers who follow tech policy. Emergent Wire covers the AI industry, and this piece sorts what the draft actually says from what advocates and opponents claim.

What Does the Great American AI Act Draft Say?

The draft pairs a federal framework with a pause on state action. As summarized in The Regulatory Review, it applies to frontier AI developers with more than $500 million in annual revenue. It requires published safety frameworks, critical incident reporting, and semi-annual third-party audits.

It would also formally establish a Center for AI Standards and Innovation inside the National Institute of Standards and Technology (NIST). That center would build on the former AI Safety Institute, develop guidelines, and evaluate domestic and foreign AI systems.

Elegant courtroom seating with leather chairs and wooden desks, exuding a sense of authority.

CASRAI's September explainer lists the bipartisan sponsors as Reps. Jay Obernolte, Lori Trahan, Suhas Subramanyam, Scott Franklin, Scott Peters, and Erin Houchin. It says the draft would federalize versions of some state frontier-safety and transparency rules.

How Would Three-Year Preemption Work?

Preemption means federal law overrides conflicting state law. Here, the override would last three years and apply to state laws that regulate how AI models are developed. Laws that govern deployment or use would be left alone.

The draft names specific California statutes. CASRAI reports that it targets AB 2013, the training-data transparency law, and parts of SB 942, the AI content watermarking law, for outright preemption. That detail matters because those laws already shape what some developers publish.

The Regulatory Review's author puts the distinction vividly, as the difference between regulating the conditions under which a weapon is built and regulating only where it can be fired. His argument is that the development layer is where many safety duties belong, so a pause there leaves a gap.

Why Do Supporters Want One National Rule?

Supporters argue that a patchwork of state laws is costly and confusing. A developer that serves users in every state may need to track dozens of different disclosure and testing rules. A single federal standard, they say, would be easier to follow and to enforce.

The executive branch has pushed the same direction. CASRAI notes that Executive Order 14365, signed December 11, 2025, directed a Justice Department task force to challenge state AI laws and instructed the administration to prepare a federal preemption bill. Congress had already rejected one version of this idea, since the Senate stripped a 10-year moratorium on state AI enforcement from a reconciliation bill by a 99-1 vote in 2025.

That history is useful context. Preemption has been tried, and it has failed more than once. The new draft is narrower, tied to a federal framework, and limited to three years, which is an attempt to answer earlier objections.

What does this mean in dollars and time? A compliance program for a large developer involves legal review, internal testing, documentation, and outside auditors. Semi-annual audits mean two external reviews every year, which is a recurring cost rather than a one-time project. Companies that already publish safety frameworks would find the transition easier than those that do not.

Who Opposes the Preemption Clause and Why?

Opposition is broad. CASRAI reports that in July 2026 more than 200 state lawmakers and labor groups, including the Association of Flight Attendants, urged Congress to reject the preemption provision. Their concern is that states have moved faster than Washington and should not be sidelined.

The Regulatory Review's author calls the federal framework serious but the preemption clause the main flaw. He says it creates a three-year gap in state governance at the development layer and undermines state experimentation. His proposed fix is a federal floor, not a ceiling, so that states can go further than the federal minimum.

There is a fair counterpoint. Federal audits and incident reporting would apply nationwide, and some states have no AI law at all. For those residents, a federal floor could add protection even with preemption attached.

Smaller developers and startups sit in a gray zone. The revenue threshold would exempt many of them from the federal audit duties, yet they would still face state deployment rules. That could favor the biggest labs, which can absorb audit costs, while giving newer entrants a lighter load. Critics and supporters read that tradeoff differently.

What Is Happening at the State Level Right Now?

States are not waiting. Colorado passed SB 26-189 on May 14, 2026, which repeals and reenacts its AI Act in a narrower form. CASRAI says it replaces the original duties with four obligations: user notification, adverse-outcome disclosure, a data-correction right, and meaningful human review. It takes effect January 1, 2027.

California is also active. CASRAI cites a Fortune report from September 14, 2026, that an AI watchdog alleged OpenAI's latest model releases may not meet SB 53's frontier-safety and transparency requirements. New York has enacted its own framework, according to The Regulatory Review.

For a broader look at where states stand, see our overview of state AI laws in 2026. It shows how quickly the map has changed.

An unrecognizable person with binary code projected, symbolizing cybersecurity and digital coding.

How Does the EU AI Act Change the Calculation?

The EU AI Act applies to any provider placing AI systems on the EU market, wherever the company is based. The Regulatory Review notes that the European Commission appointed a 60-member scientific panel and a 174-member advisory forum to support enforcement. A U.S. company that meets only a minimum federal standard would still face full EU requirements.

That shapes strategy. A company can adopt a lighter U.S.-specific posture during any preemption window, or harmonize globally to the higher EU bar. Many large developers will likely choose the second path, because maintaining two compliance programs is expensive.

Our explainer on the EU AI Act in 2026 covers the obligations and timelines in more detail.

Is the Draft Likely to Become Law?

Not soon. CASRAI describes the draft as stalled since June, and notes that House Speaker Mike Johnson said in mid-September that Congress should not panic into emergency AI regulation. His remarks concerned pausing frontier AI development, not state preemption, but they signal low appetite for rushed legislation.

The draft also has to be formally introduced before it can move. Until then, it is a negotiating text. Sponsors may narrow it, shorten the preemption window, or add a federal floor to win votes.

Enforcement is a second open question. A federal framework is only as strong as the agency that runs it. Emergent Wire's editors will watch whether the Justice Department task force created under the executive order files and wins a case against a state AI law, since a court win could change the leverage in Congress.

A detailed close-up of a Lady Justice statue holding scales, symbolizing law and justice.

There is also a practical question about timing. Three years is a long time in AI. Model release cycles now run in months, and a rule written around today's systems may not fit the ones that arrive in 2028. Our look at AI model release cycles in 2026 shows how fast the target moves.

What Should AI Companies and Readers Watch?

Companies should plan for several outcomes instead of betting on one. Build compliance around the strictest rules you already face, and treat any federal relief as a bonus. Document safety frameworks and incident processes now, since the draft would require them anyway.

Readers can follow four signals:

  1. Whether the draft is filed as a formal House bill.
  2. Whether the three-year preemption window survives or shrinks.
  3. How Colorado's attorney general writes the rules for the January 1, 2027 start date.
  4. How regulators resolve the California SB 53 dispute over frontier releases.

Our look at the FTC's civil investigative demands to OpenAI and Anthropic shows that federal agencies are already active, even without a new law.

Demonstrators in Prague holding signs advocating for political change and democracy.

One last point is worth stating plainly. Nothing in the draft changes what companies must do today. Existing state laws remain in force, the EU AI Act continues to apply, and agencies keep their current powers. Planning around a bill that may never pass is a risk, so treat the draft as a signal of direction rather than a deadline.

Conclusion

The Great American AI Act draft tries to trade a serious federal framework for a three-year pause on state laws about model development. Supporters see clarity, and opponents see a gap in oversight. Both sides agree that the status quo is messy.

Your next step is to track whether the draft becomes a bill, and to keep compliance work tied to the strictest rule you already face. Emergent Wire will update this analysis as the text changes.

What is the Great American AI Act?
It is a bipartisan House discussion draft released on June 4, 2026, and not yet a filed bill. It would create federal duties for large frontier AI developers, including published safety frameworks, incident reporting, and third-party audits, and it would preempt state laws on how AI models are developed for three years.
Would the draft stop states from regulating AI entirely?
No. According to the draft as described by The Regulatory Review and CASRAI, the three-year preemption covers state laws regulating how AI models are developed. States would keep authority over how AI is deployed and used, so rules on hiring tools, chatbots, or healthcare disclosures would generally remain open to the states.
Which companies would the federal rules cover?
The framework described by The Regulatory Review applies to frontier AI developers with more than $500 million in annual revenue. Those companies would need to publish safety frameworks, report critical incidents, and undergo semi-annual third-party audits. Smaller developers would fall outside the main requirements, though they could still face state deployment rules.
Is the bill likely to pass soon?
It does not look likely in the near term. CASRAI describes the draft as stalled since June, more than 200 state lawmakers and labor groups urged Congress in July to reject the preemption provision, and House Speaker Mike Johnson has signaled he does not want to rush emergency AI regulation. Passage depends on the draft being formally introduced first.

Written by

Tobias Reyes

AI Industry & Policy Analyst

12+ years experience

Tobias Reyes spent ten years as a technology strategy consultant, most of that time advising Fortune 500 clients on enterprise software adoption before AI became the whole conversation. He joined Emergent Wire to cover the industry from the incentive layer up: who actually pays for frontier model access, how compute costs shape product decisions, and where policy is likely to reshape the competitive landscape next. He is allergic to roadmap-slide analysis and prefers to start from unit economics and public filings whenever they exist.

Credentials

  • Former technology strategy consultant (10 years, enterprise clients)
  • MBA, focus on technology and operations strategy
  • Regular commentator on AI industry economics and policy