The Great American AI Act Bets on Federal Preemption. The Last Three Bets Lost.

The Core Bet

The compliance fragmentation problem in US AI is real and getting worse. The question is whether Congress is actually ready to solve it, or whether this week's bipartisan House draft is the fourth attempt in a row to do something that keeps dying on arrival.

Reps. Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.) rolled out a discussion draft Thursday that would grant the federal government regulatory control over various aspects of AI while preempting state regulation for a three-year period. The Great American Artificial Intelligence Act of 2026 spans 269 pages and represents the most substantive bipartisan legislative text on AI governance to reach public discussion draft stage in this Congress.

But context matters. Congress has considered two significant legislative attempts to preempt state AI regulations and declined to enact either. A provision in the One Big Beautiful Bill Act would have preempted state AI regulations for 10 years; after significant media scrutiny and bipartisan opposition, the Senate voted 99-1 to strip it. Congress then declined to enact a similar moratorium through the 2025 National Defense Authorization Act. The December 2025 Trump executive order attempting preemption through the executive branch raised immediate constitutional questions about statutory authority. Three previous tries have all failed.

What the Bill Actually Does

The preemption mechanism here is narrower than its predecessors—a political strength that masks operational weakness.

The bill would preempt state laws and regulations "specifically regulating the development" of an AI model, with a three-year sunset. Preemption would not apply to laws related to the use or deployment of AI models. That distinction matters operationally. California's frontier-model disclosure law requiring developers to publicly post training data summaries gets swept away. Colorado's requirement that deployers conduct impact assessments before making consequential housing or employment decisions survives.

An accompanying document released by Trahan's office said California's AB 2013, which requires model developers to publicly post high-level summaries of their training data, would be preempted, along with a portion of California SB 942 related to content watermarking.

The affirmative side: the bill would require large frontier developers with more than $500 million in gross revenue for the previous calendar year to establish public frontier AI frameworks. Those frameworks must assess whether a model could pose catastrophic risk. This federal floor replaces state-level developer rules, at least for the largest players.

The draft creates four pillars: establishing frontier AI model governance, collecting insight into changes within the US workforce landscape, fortifying cybersecurity postures, and spurring new AI research and development.

The State Landscape the Bill Is Trying to Simplify

The backdrop here is genuinely complex. In 2025 alone, all 50 states, Washington D.C., Puerto Rico, and the US Virgin Islands introduced AI-related legislation, with 38 states adopting or enacting such laws. That is not a patchwork—it is a thicket.

The most operationally significant divergences are in high-risk AI. Colorado SB 205 created one of the first comprehensive state AI regimes, regulating "high-risk artificial intelligence systems" used in "consequential decisions," imposing broad obligations on developers and deployers related to risk management, impact assessments, consumer disclosures, and reporting to the Colorado attorney general. Colorado's enforcement date has already shifted from February 1, 2026, to June 30, 2026, after a special legislative session in August 2025, with a March 2026 working group draft pushing it further to January 1, 2027.

New York moved differently. Gov. Kathy Hochul signed the RAISE Act in December 2025, then signed amendments on March 27, 2026, shifting it toward a transparency and reporting-based framework that mirrors California's approach. New York's revised law carries civil penalties up to $1 million for a first violation and up to $3 million for subsequent violations, with a 72-hour incident reporting timeline.

For a multistate deployer, the friction is immediate. Using AI in hiring decisions means navigating overlapping requirements from NYC Local Law 144, Illinois's video interview rules, Maryland and New Jersey hiring restrictions, California's civil rights department regulations, and federal anti-discrimination statutes. That is five separate compliance regimes for one use case.

Why the Preemption Clause Is the Weak Link

The development-only preemption scope solves the compliance problem for frontier model builders. It does essentially nothing for the enterprise deployer—where the operational pain actually lives. The $500 million revenue threshold covers OpenAI, Google DeepMind, Anthropic, and Meta. The majority of companies running AI systems in consequential decision-making contexts fall outside it.

The deeper problem is enforcement. While federal preemption eliminates the need to track multiple state regimes, organizations should not assume it translates to reduced compliance expenditure; it will more likely signal a shift in where those resources are directed. A federal framework with weak enforcement mechanisms and vague risk-management definitions becomes compliance theater rather than compliance simplification.

Companies should continue to comply with applicable state AI laws because an executive order itself does not, and cannot, overturn existing state law. Until relevant legal challenges are resolved, state laws remain enforceable, and companies could face potential penalties for noncompliance. The same holds for this discussion draft until enactment and constitutional survival.

The Political Math

Bipartisanship in sponsorship exists but has limits. Reps. Suhas Subramanyam (D-Va.), Scott Franklin (R-Fla.), Scott Peters (D-Calif.), and Erin Houchin (R-Ind.) joined Obernolte and Trahan in releasing the draft. That signals meaningful cross-party buy-in at the sponsor level.

The House Democratic Commission on AI and the Innovation Economy, chaired by Reps. Valerie Foushee, Ted Lieu, and Josh Gottheimer, did not support the current discussion draft, saying it "does not meet the enormity of the moment." That matters: those members sit on committees of jurisdiction.

While key Republican lawmakers remain supportive of federal preemption, significant Democratic opposition, particularly among members serving on panels of jurisdiction, may complicate the path forward, especially given the razor-thin GOP majority in the House.

The safety community is also skeptical. Brad Carson, president of Americans for Responsible Innovation, called the preemption of state laws a "generational mistake," saying the bill turns "the current floor on state AI legislation into a federal ceiling."

What to Watch

This is a discussion draft, not legislation. The distance to law is substantial.

1. Committee markup signals. A markup within 60 days means momentum. Stalling through summer suggests the effort is stalling.

2. How the preemption clause evolves. The development-only scope is politically calibrated but operationally thin. Watch whether deployer-side preemption gets added under industry pressure, which would expand the scope materially and likely kill Democratic support.

3. Whether the $500 million revenue threshold holds. Lowering it brings more companies under the federal framework but raises compliance costs for mid-market AI companies. Raising it narrows the bill to a handful of frontier labs and weakens the preemption case.

4. State AG responses. States with AI laws will likely attempt to block unilateral federal preemption, which could further complicate implementation. Watch for coordinated state AG letters or pre-litigation filings if the bill advances.

5. Enforcement agency designation. The bill's risk-management requirements are only as good as the agency enforcing them. Whether that is NIST, FTC, or a new body determines how much the federal standard actually bites.

The preemption debate has now run through budget reconciliation, a defense authorization bill, an executive order, and a White House legislative framework. None became law. This draft is structurally better designed than its predecessors, but the enforcement question and the deployer-side gap remain unresolved. Treat it as directional signal, not operational certainty.