Submitted statement to the House Judiciary Subcommittee on Courts, Intellectual Property, Artificial Intelligence, and the Internet, at a hearing held September 18, 2025 before Chairman Issa and Ranking Member Johnson. Frazier is identified as "the inaugural AI Innovation and Law Fellow at The University of Texas School of Law."
The testimony is a constitutional argument for federal primacy over frontier-model development, built from three founding-era principles rather than from a policy assessment of any particular state statute. Its operative conclusion is a division of labour between levels of government, not a general preemption of state AI law.
The hearing's framing
The hearing's title — "AI at a Crossroads: A Nationwide Strategy or Californication?" — states the preemption question as a binary and signals the majority's framing: state-led regulation, California's in particular, as a de facto national standard set without national deliberation.
That framing is the counterpart to the reverse federalism argument advanced by OpenAI, which reads the same state activity as a consensus Congress should adopt rather than as an imposition it should displace. The two positions agree that states have set the operative rules and disagree on whether that is a starting point or a problem.
The three principles
Federal responsibility for national stability. Frazier argues that "the federal government alone is responsible for matters that implicate the economic and political stability of the country, while states maintain considerable discretion to address local concerns," and that "it cannot be the case that the absence of an affirmative federal response to such a national issue invites or permits state action." He treats the alternative — that "each state has the authority to set for itself the limit of its regulatory powers" — as an invitation to "the serial testing of state authority" which, "if replicated by 50 states, will result in jurisdictional squabbles at a minimum and, potentially, national discord." The historical argument runs through the transition from the Articles of Confederation, citing Hamilton in Federalist 22 on experiments that undermine uniformity in trade and diplomacy, and Madison in Federalist 42 on "different regulations of the different states" undermining economic competitiveness.
The territorial limit on state authority. The second principle is that "the extensive authorities reserved to each state end at their respective borders," resting on the equal sovereignty of the states — "equal in power, dignity, and authority" (Coyle v. Smith, 221 U.S. 559, 567 (1911)) — and on Kansas v. Colorado, 206 U.S. 46, 97 (1907): "[o]ne cardinal rule, underlying all the relations of the States to each other, is that of equality of right." Frazier argues that "[p]olitical clout, economic might, and sheer population does not grant any state the authority to step into the shoes of the federal government," illustrating the point with Virginia's roughly 20 percent share of the founding-era population and with the colonial "king of New York" levying entrance and clearance fees on transiting vessels. "Whether a state is the fourth largest economy in the world or the one hundredth and fourth should have no bearing on its authority to shape the lives of Americans beyond its borders."
Individual liberty and consent. The third principle is "that the ultimate authority in our constitutional system rests with the people," and that consent is an aspect of the social contract. Under the Articles of Confederation, Frazier argues, citizens "found their lives being dictated by decisions made in other states." Applied to AI: "Americans may be able to move as freely as they'd like, but they will still find themselves under the thumbs of state legislators over which they have no control."
The Pork Producers distinction
The testimony's most specific legal move is to distinguish National Pork Producers Council v. Ross, 598 U.S. 356 (2023), which upheld a California prohibition on the sale of pork raised under certain conditions against a dormant Commerce Clause challenge. Frazier argues the case does not extend to frontier-model regulation because the California measure "did not compel out-of-state producers to wholly change their entire production process nor alter the nature of the underlying product."
The distinction turns on the indivisibility of a training run: "Whereas a pig is a pig; a model trained to comply with state A's requirements will differ from one trained under state B's mandates. Labs cannot afford to conduct 50 different training runs, each of which may last for several months and involves an incredible amount of data and compute." Any change made for one state's rules therefore "will impact the model made available to the rest of the country" and, he adds, in some cases to billions of people elsewhere. He names California, Colorado and New York as the jurisdictions whose laws may force such changes, and states the resulting cost as one borne by non-residents: "progress delayed is progress denied."
The argument is a technical premise carrying a constitutional conclusion, and its force depends on that premise holding. It is the same indivisibility claim that underlies the industry position on a state patchwork, here put to a doctrinal use rather than a policy one. Frazier cites Matt Perault's "Regulate AI Use, Not AI Development" and Epoch AI's training-cost work in support. See State-Level AI Regulation and AI Federalism.
What the testimony concedes to states
The statement is not an argument that states should be displaced from AI regulation. It says states "can and should implement regulations responsive to the demands of their political community and demonstrated to achieve those ends," and proposes procedural discipline for finding the line: "retrospective review requirements, sunset clauses, and frequent independent evaluations." It also identifies interventions it treats as carrying "little to no odds of interfering with the actual AI tools" — "increased training for end users, AI literacy programs, and procurement standards that align with the state's values."
The closing states the division of labour directly: "the development of frontier AI models is a national undertaking; the uses of those systems within a state are a proper subject of deployment rules tailored to local concerns." States "should be empowered—indeed, encouraged—to police unfair and deceptive practices, to adopt procurement standards, and to specify disclosure obligations in particular contexts," while "Congress should take up the regulatory tasks with nationwide consequences." Frazier frames the question before the subcommittee as "not whether to protect Americans from the genuine risks posed by advanced AI, but how to do so while preserving the constitutional design."
Provenance
Full text extracted from the University of Texas School of Law's copy of the submitted statement, a six-page PDF with 45 footnotes (Source: law.utexas.edu). The statement is a submitted written testimony; no transcript of oral testimony or of the subcommittee's questioning is held here. Frazier opens by referring to earlier testimony before the same subcommittee by Benjamin Jensen, that "the nation that leads in AI will shape the future."
Relationships
- contradicts: Reverse Federalism — the competing reading of the same state-level activity, which treats state law as a consensus for Congress to adopt
- supports: AI Federalism — a constitutional rather than policy argument for federal primacy over model development
- related: State-Level AI Regulation, California SB 53, Colorado AI Act (SB 24-205), New York RAISE Act — the state measures the testimony names or implies
- related: Frontier AI Governance, Democratic Governance of Frontier AI: A blueprint for a federal framework (OpenAI, June 2026)