The Evidence Pack: Inside the 'Great American AI Act' and the Push to Freeze State AI Laws

The regulatory landscape for artificial intelligence in the United States is hurtling toward a constitutional and commercial collision. A bipartisan coalition in the U.S. Senate has released the draft text of the "Great American AI Act," a sweeping legislative package designed to establish the first comprehensive federal framework for artificial intelligence. The bill proposes mandatory safety audits for frontier models, a new federal oversight office, and strict export controls. But the core of the legislation—and the source of immediate, intense controversy—is a provision that would explicitly freeze all state-level AI regulations for a period of three years.[1]

This "evidence pack" examines the claims, the data, and the legal uncertainties surrounding this proposed preemption. The tension fundamentally pits the tech industry's demand for a unified national market against state governments' desires to protect local consumers from algorithmic harm. Over the past two years, in the absence of federal action, states like California, Illinois, and Connecticut have aggressively filled the void, passing a patchwork of laws governing everything from deepfakes in elections to algorithmic bias in hiring and housing.[2][5]

The draft legislation, spanning nearly 400 pages, seeks to centralize authority. It establishes the Federal AI Oversight Office (FAIO) within the Department of Commerce, armed with a proposed $500 million annual budget. The FAIO would be tasked with maintaining a national registry of frontier models—defined as those trained using more than 10^26 floating-point operations—and enforcing mandatory pre-deployment safety testing. Crucially, Section 402 of the draft asserts that these federal standards will supersede any state statute regulating the training, deployment, or auditing of generative AI models.

The primary claim driving the push for federal preemption is that state-level fragmentation is actively stifling American innovation. Proponents argue that requiring AI developers to comply with 50 different, often contradictory, regulatory regimes creates an impossible burden, particularly for open-source developers and startups. The tech industry has lobbied heavily for this harmonization, arguing that AI models, by their nature, do not respect state borders and cannot be easily geofenced to comply with localized rules.[2][3]

Economic analyses provide some backing for this claim. A recent report from the Brookings Institution modeled the compliance costs of the current trajectory, estimating that a fully fragmented state regulatory environment could increase legal and operational costs for mid-sized AI enterprises by up to 40%. The report concluded that a unified federal standard, even a stringent one, would ultimately accelerate enterprise AI adoption by removing legal ambiguity and providing a stable environment for long-term capital investment.

However, the counter-evidence is equally forceful, rooted in the immediate loss of consumer protections. Opponents of the bill argue that the federal government moves too slowly to regulate a technology evolving at an exponential rate. They point out that the proposed three-year freeze is designed to give the newly formed FAIO time to draft and implement its rules. During that window, existing state laws would be suspended, potentially creating a regulatory vacuum.[5]

The Electronic Frontier Foundation (EFF) has sharply criticized the draft, publishing an analysis that outlines the specific harms of a three-year freeze. According to the EFF, suspending state laws would immediately void critical protections against algorithmic discrimination in healthcare and employment that are currently active in several states. The organization argues that the bill sacrifices the safety of vulnerable populations to ensure frictionless deployment for major technology companies.

The mechanics of the freeze are legally complex. If enacted, the bill would immediately preempt 14 active state laws specifically targeting AI. Enforcement would revert to existing, broader federal civil rights and consumer protection statutes, which many legal scholars argue are ill-equipped to handle the nuances of generative AI and autonomous agents. State Attorneys General, who have been the primary enforcers of tech regulation in recent years, would find their authority severely curtailed.[3][4]

This dynamic has triggered a fierce backlash from state capitals. Several State Attorneys General are already preparing the groundwork for a constitutional challenge should the bill pass. They argue that the preemption clause violates the Tenth Amendment by improperly commandeering state regulatory authority. The states maintain that consumer protection has historically been a shared jurisdiction, and that states must retain the right to act as "laboratories of democracy" in the face of novel technological threats.[4]

Legal analysts at Bloomberg Law note that the federal government's position is legally precarious but historically grounded. Congress possesses broad authority under the Commerce Clause to regulate technologies that operate across state lines. However, explicitly freezing state consumer protection laws without an immediate, fully operational federal replacement is an aggressive use of that power, likely to trigger years of litigation that could reach the Supreme Court.[4]

The political reality in Washington suggests the bill has significant momentum. The bipartisan nature of the draft indicates a growing consensus among federal lawmakers that the U.S. cannot afford a fragmented domestic market while competing globally, particularly with China's heavily centralized AI infrastructure and the European Union's unified AI Act. The Senate Commerce Committee has scheduled a series of high-profile hearings for next month to debate the preemption clause.[1]

Yet, the uncertainty remains high. The evidence pack reveals a stark trade-off: the Great American AI Act promises the regulatory clarity necessary for the U.S. to maintain its lead in frontier AI development, but it demands the sacrifice of localized, immediate safeguards. As the bill moves from draft to formal introduction, the debate will center not on whether AI should be regulated, but on who holds the power to write the rules.[5]