On December 11, 2025, President Trump signed Executive Order 14365 — Ensuring a National Policy Framework for Artificial Intelligence (published Dec. 16; 90 FR 58,499). Eight days later, New York Governor Hochul signed the RAISE Act, and that same week 23 state attorneys general filed against federal preemption at the FCC. A federal move to override state AI law and a state push to regulate frontier AI collided inside eight days — the visible start of the US AI constitutional dispute.
This is not a partisan clash but a constitutional question: can the executive preempt state law through an executive order, agency rulemaking, and DOJ litigation alone, with no federal statute? The United States has no federal AI statute, so every attempt is post-statutory — an executive preemption strategy to neutralize 50 states' legislation through executive power. In the absence of a federal AI statute, the preemption the Trump administration is attempting is being tested at the frontline of three constitutional doctrines at once — the Supremacy Clause, the Dormant Commerce Clause, and the First Amendment — and the outcome will determine the structure of US AI governance.
Why December 2025
Three layers converged. (1) A state-law explosion — 1,000-plus AI bills across the 50 states in 2025, 38 enacted: Colorado SB 24-205 (omnibus), California TFAIA and NY RAISE ($500M frontier model), Texas TRAIGA (intent-based), Utah (disclosure). (2) Two congressional failures — a May 2025 budget-bill moratorium, opposed by 40 AGs, stripped 99–1 in July; a November NDAA preemption amendment, opposed by a 36-AG letter (reportedly led by NC AG Jeff Jackson, with UT's Brown, NH's Formella, NY's James, through NAAG), dropped from the final bill. (3) The executive workaround — blocked twice in Congress, the administration turned to executive order, agencies, and litigation. That is why post-statutory executive preemption became the frontier strategy.
EO 14365's four weapons
The order is an agency directive, not legislation, and does not preempt directly. (1) The Commerce evaluation (Sec. 4) — assess onerous state laws by March 11, 2026; deadline passed, unpublished as of May 2026, since a formal target list becomes a justiciable instrument. (2) The FCC proceeding (Sec. 6) — within 90 days of the evaluation, relying on § 253, but AI is an information service, not telecommunications; Commissioner Anna Gomez herself called the authority "dubious," with CRS and Public Knowledge agreeing. (3) The FTC deceptive-practices reading (Sec. 7) — that a state law forcing alterations to truthful AI output is preempted under FTC Act § 5; combined with EO 14319 (Preventing Woke AI), it casts state AI law as compelled viewpoint discrimination. (4) The BEAD condition (Sec. 5) — onerous-law states lose non-deployment funds, near the NFIB v. Sebelius (2012) ten-percent coercion benchmark and likely failing the South Dakota v. Dole (1987) relatedness prong (BEAD funds broadband, not AI). All four deadlines passed with no action — likely deliberate. Adjuncts: the AI Litigation Task Force (Jan. 9, 2026 Bondi memo; reportedly including the Solicitor General's office; David Sacks consulted) and the National AI Legislative Framework (Mar. 20, 2026).
Three constitutional frontlines
Supremacy Clause. EO 14365's Section 9(c) concedes it is non-justiciable on its own. Under Youngstown (1952), with no AI statute and two congressional refusals readable as an implied will against preemption, executive action sits at Tier 3 — the lowest ebb.
Dormant Commerce Clause. American Libraries Ass'n v. Pataki (S.D.N.Y. 1997): cyberspace has no borders, so a single state's rule has nationwide effect and requires uniform national treatment. National Pork Producers Council v. Ross, 598 U.S. 356 (2023): extraterritorial impact alone is not enough — the doctrine narrowed. Whether AI, like the 1997 internet, cannot be partitioned by geography is the frontier question; the DOJ dropped this weaker claim and concentrated on Equal Protection.
First Amendment. Moody v. NetChoice, 603 U.S. 707 (2024): a 9–0 vacate/remand, but Kagan's opinion was joined in full by only Roberts, Sotomayor, Kavanaugh, and Barrett (Jackson joining Parts I, II, III-A), with four concurrences — agreement that editorial discretion is protected speech, divergence on application. xAI v. Weiser (Count 1) quotes Moody at 741–42 to frame AI generation as editorial discretion; Paul Clement, who argued Moody for NetChoice, argues both xAI cases. Langvardt & Rozenshtein, Beyond the Editorial Analogy, 6 J. Free Speech L. 1 (2025), note the concurring Justices were "unmoved" by the claim that computers are inherently expressive. Judge Bernal's first ruling (xAI v. Bonta, PI denied Mar. 4, 2026; 9th Cir., docket no. reported as 26-1591) treated training-data disclosure as factual commercial speech under lower scrutiny; the behavioral-regulation space (Colorado) is open.
A new institutional actor
The cumulative letters — 40 (May), 36 (November), 23 (December 2025) — institutionalized a bipartisan coalition, NAAG as conduit, Republican AGs (reportedly Iowa's Bird, NH's Formella, UT's Brown) signing alongside Democrats. The state versus federal framing of the 1987 Dole era has evolved into agency versus AG coalition.
The tests ahead
Four inflection points: Colorado's June 30, 2026 effective date; publication of the Commerce evaluation (inviting D.D.C. declaratory actions); an FCC declaratory ruling (§ 253's reach, a 23-AG challenge); and an NFIB-based BEAD challenge. That three doctrines are tested at once is the case's richness. After the DOJ intervention, xAI v. Weiser may be the first substantial federal ruling on a state AI law's constitutionality — the role Allen v. Perlmutter (the AI-copyright case on the human-authorship requirement) played in copyright, xAI v. Weiser may play in AI preemption.
Fifty states built their own regulation while none was preempted; the executive is trying to neutralize all fifty without a statute; the AGs are blocking it; and the courts are applying three constitutional doctrines at once. The answer arrives, in part, on Colorado's June 30 effective date; the rest is decided in the 18 months that follow. The preemption has not happened yet. It is trying to happen. It is being blocked. That process is, right now, the richest frontier dispute in American constitutional law.