Executive Order 14365 and a new DOJ task force are trying to clear the 50-state patchwork. The catch: an executive order cannot preempt state law on its own.
Can an executive order preempt state AI laws?
No. An executive order cannot preempt state law on its own, because under the Supremacy Clause only a federal statute (or a court reading one) can override a validly enacted state statute. Executive Order 14365 directs federal agencies to attack state AI laws in court and through funding leverage, but it does not, and constitutionally cannot, repeal them. As one client alert from Crowell and Moring put it, “federal preemption by executive decree, absent a clear congressional delegation of powers, is not a generally accepted practice under the U.S. Constitution.”
This distinction is the whole story of the 2026 state AI law preemption fight. The White House signed EO 14365 “Ensuring a National Policy Framework for Artificial Intelligence” on December 11, 2025, declaring a policy of sustaining U.S. AI dominance through a single, minimally burdensome national framework. But declaring a policy is not the same as having the legal power to impose it. The order itself acknowledges it can only act “to the maximum extent allowed by Federal law” — an open admission that existing law sets the ceiling.
Preemption normally comes in two flavors: express preemption, where Congress writes in a statute that it is displacing state law, and conflict or field preemption, where courts infer that federal law leaves no room for the states. Both require an act of Congress. The executive branch has no freestanding power to manufacture either one. That is why the substance of EO 14365 is not preemption itself but a toolkit of indirect pressure: litigation, funding conditions, and a legislative ask to Congress.
So the honest answer for builders is that the order changes the political weather without changing your immediate legal obligations. A Colorado or California statute on the books in 2026 remains enforceable unless and until a court enjoins it or Congress passes a preempting law. None of that has happened yet for the major state AI statutes.
This article is journalism, not legal counsel. The preemption questions here are genuinely unsettled and will likely take years of litigation to resolve. Talk to qualified counsel before making any compliance decision based on EO 14365 or the DOJ task force.
What does Executive Order 14365 actually do?
EO 14365 builds a federal pressure campaign with three levers: a DOJ litigation task force, agency reviews that name “burdensome” state laws, and funding conditions tied to the BEAD broadband program. None of the three repeals a state law; each is a way to make state AI regulation more costly to keep on the books.
The headline lever is the DOJ AI Litigation Task Force. Attorney General Pam Bondi stood it up in early January 2026, with CBS News reporting the announcement on January 9 and the task force’s mandate beginning January 10. Its sole job is to challenge state AI laws the administration deems inconsistent with national policy, consulting White House AI and crypto czar David Sacks on which laws to target. The legal theories are the Dormant Commerce Clause (that a state law unduly burdens interstate commerce), conflict preemption by existing federal regulation, and First Amendment objections to mandated AI disclosures.
The second lever is agency naming-and-shaming. The order set a March 11, 2026 deadline for the Commerce Department and NTIA to publish an evaluation identifying state laws considered excessively burdensome, with the FTC directed by the same date to weigh treating state-mandated bias mitigation as a deceptive trade practice. The third lever is money: the order ties eligibility for remaining funds in the roughly $42 billion BEAD broadband program to whether states maintain AI laws Washington dislikes.
Finally, EO 14365 instructed advisers including David Sacks and Michael Kratsios to draft a uniform federal framework that Congress could pass. That recommendation became the White House National Policy Framework for AI, released March 20, 2026, which urges Congress to broadly preempt state AI laws. The framework is the tell: the administration is asking Congress to do the one thing the executive cannot do alone.
| Date | Event | Significance |
|---|---|---|
| Jul 1, 2025 | Senate strips 10-year AI moratorium from One Big Beautiful Bill Act, 99-1 | Congress declines to preempt state AI laws via reconciliation |
| Dec 7, 2025 | FY2026 NDAA text released without AI preemption language | Second congressional rejection of a federal preemption vehicle |
| Dec 11, 2025 | President signs EO 14365 “Ensuring a National Policy Framework for AI” | Executive branch pivots to litigation and funding pressure |
| Jan 9-10, 2026 | DOJ AI Litigation Task Force announced and stood up under AG Bondi | Federal enforcement arm to challenge state laws in court |
| Mar 11, 2026 | Commerce/NTIA and FTC deadline to flag “burdensome” state laws | Agencies build the target list for litigation and funding leverage |
| Mar 20, 2026 | White House releases National Policy Framework for AI | Asks Congress for broad preemption the EO cannot deliver alone |
| Apr 27, 2026 | Federal court stays Colorado AI Act enforcement; DOJ intervened | First major courtroom collision over a state AI statute |
Why couldn’t Congress just preempt state AI laws?
Congress has now declined to preempt state AI laws twice in under a year, which is exactly why the administration fell back on an executive order. Both attempts failed in bipartisan fashion, leaving the executive branch without the federal statute that real preemption requires.
The first attempt was the proposed 10-year moratorium on state AI regulation tucked into the House version of the One Big Beautiful Bill Act. On July 1, 2025, during a marathon vote-a-rama, the Senate voted 99-1 to strip it out, after a compromise that would have shortened the ban to five years collapsed under criticism from both parties. As Reason reported, the lone holdout could not save a provision that drew opposition from senators ranging from Maria Cantwell to Marsha Blackburn to Ed Markey and Susan Collins.
The second attempt was an effort to reinsert AI preemption into the FY2026 National Defense Authorization Act. StateScoop reported that when the NDAA conference text emerged on December 7, 2025, the preemption language was gone, opposed by Armed Services leaders and by a striking cross-section of Republican governors and senators — Ron DeSantis, Sarah Huckabee Sanders, and Josh Hawley among them. EO 14365 followed four days later.
The pattern matters for builders trying to forecast risk. Twice, the most plausible legislative vehicles for nationwide preemption died, and the opposition was not partisan but federalist — a coalition that values states’ ability to regulate emerging tech. That makes a near-term federal statute, the only thing that could truly clear the patchwork, look politically difficult even with a sympathetic White House.
“Twice in under a year Congress declined to preempt state AI laws. The executive order is what you reach for when the statute you actually need will not pass.”
Surya Koritala, founder of Cyntr
What did Colorado and California do?
Colorado delayed and is reworking its first-in-the-nation AI Act while fighting a court challenge, and California pressed ahead with frontier-model transparency rules and is publicly opposing preemption. The two states bracket the spectrum of how state capitals are responding to federal pressure.
Colorado’s AI Act (SB 24-205), the broad “algorithmic discrimination” law that EO 14365 singles out by name, has been repeatedly pushed back. A 2025 special session produced SB 25B-004, moving the effective date to June 30, 2026. Then litigation hit: Troutman reports that on April 24, 2026 the Colorado Attorney General filed a joint motion not to enforce the law pending interpretive rulemaking, and a federal court granted it on April 27, 2026, with the U.S. Department of Justice having intervened to argue the law should be enjoined. Colorado’s pioneering statute is, for now, on ice.
California took the opposite tack. Governor Newsom signed SB 53, the Transparency in Frontier Artificial Intelligence Act, on September 29, 2025, and it took effect January 1, 2026. It targets only the largest developers — foundation models trained above 10^26 FLOPs — and requires published transparency reports, catastrophic-risk assessments, critical-incident reporting to the state within 15 days (24 hours if there is imminent danger), and whistleblower protections, with civil penalties up to $1 million per violation. Attorney General Rob Bonta has separately written to Congress opposing AI preemption, framing it as an assault on the federalist system.
The contrast is instructive. Colorado’s troubles came less from EO 14365 and more from its own implementation difficulty plus a private lawsuit the DOJ joined. California’s transparency-first design — narrow scope, disclosure rather than liability — has so far proven more durable and more defensible. For builders, the live obligations in 2026 are concrete: SB 53’s reporting duties apply now, and Colorado’s law could switch back on after its rulemaking concludes.
EO 14365 has not suspended a single state statute. SB 53’s transparency and incident-reporting duties are enforceable in California today, and the Colorado AI Act is delayed, not dead. Build to the laws that are actually in effect, not to the executive branch’s stated goal.
How strong is the federal legal position on state AI law preemption?
A pressure campaign, not a preemption
The federal legal position is genuinely contested, and several of the EO’s mechanisms face serious constitutional headwinds. Lawyers across the spectrum agree the order is more litigation strategy than settled law, and that state statutes stay in force throughout the fight.
Start with the Supremacy Clause. It lets federal statutes preempt state law, but an executive order is not a statute. Courts are even more reluctant to find preemption based on agency regulations than on legislation, which undercuts the conflict-preemption theory the DOJ task force plans to press. Without a federal AI statute to point to, the strongest preemption arguments have nothing to attach to.
Next, the anti-commandeering doctrine. The Tenth Amendment, as the Supreme Court has read it, bars the federal government from simply ordering states not to enact or enforce laws regulating private conduct. An executive order that tries to forbid state AI legislation runs straight into that wall. The BEAD funding lever faces a parallel problem under the Spending Clause, which limits how coercive federal funding conditions can be — conditioning $42 billion in broadband money on repealing AI laws is the kind of leverage that invites a constitutional challenge.
The Dormant Commerce Clause is the administration’s most plausible theory, since a genuinely extraterritorial or unduly burdensome state law can be struck down on that ground. But it is a case-by-case argument, not a blanket override, and narrowly drawn transparency laws like SB 53 are harder to characterize as protectionist burdens on interstate commerce. The practical upshot, as Crowell and Moring noted, is that “state laws will likely remain in effect” while these questions grind through the courts — possibly all the way to the Supreme Court.
DOJ task-force litigation (Dormant Commerce Clause)
Best for: Challenging broad, arguably extraterritorial state mandates
What works
Watch out for
Conflict preemption by federal regulation
Best for: Almost nothing until Congress acts
What works
Watch out for
BEAD funding conditions (Spending Clause)
Best for: Political pressure on state legislatures
What works
Watch out for
Builder’s take
I build Cyntr, an agent-orchestration runtime, and Loomfeed, so I treat the regulatory map as a hard input to architecture, not a press-release sideshow. The temptation in 2026 is to read the headlines about EO 14365 and assume the states have been neutralized. That is a compliance trap, because the executive branch cannot actually erase a state statute by itself.
- Build to the strictest live state regime, not to the executive order. Colorado and California statutes are still law; the EO is a litigation and lobbying strategy, not a repeal.
- Instrument for transparency-style obligations now. SB 53 incident reporting and Colorado-style impact disclosures both push toward model documentation you should be generating anyway for your own observability.
- Treat the BEAD funding threat as orthogonal to your product. Conditioning broadband money is a fight between Washington and state capitals; it does not change what your agent must do to ship in California today.
- Assume multi-year uncertainty. Real preemption needs a federal statute or a Supreme Court ruling, and neither exists yet, so design for a patchwork that survives at least through 2027.
Frequently asked questions
No. An executive order cannot preempt state law on its own. Under the Supremacy Clause, only a federal statute (or a court interpreting one) can override a validly enacted state law. EO 14365 directs agencies to challenge state AI laws through litigation and funding conditions, but it does not repeal them, and the order itself only claims to act ‘to the maximum extent allowed by Federal law.’
It is a unit at the Department of Justice, stood up under Attorney General Pam Bondi in early January 2026, whose mandate is to challenge state AI laws the administration considers inconsistent with national policy. It plans to use the Dormant Commerce Clause, conflict preemption, and First Amendment theories, and consults White House AI czar David Sacks on which state laws to target.
Congress declined twice in under a year. The Senate stripped a proposed 10-year state AI moratorium from the One Big Beautiful Bill Act in a 99-1 vote on July 1, 2025, and AI preemption language was left out of the FY2026 NDAA when its text emerged on December 7, 2025. Opposition was bipartisan and federalist, including several Republican governors and senators.
It is delayed, not repealed. The Colorado AI Act (SB 24-205) had its effective date pushed to June 30, 2026 via SB 25B-004, and on April 27, 2026 a federal court stayed enforcement after the state agreed not to enforce pending interpretive rulemaking, with the DOJ intervening in the litigation. The law could resume once rulemaking concludes.
California’s SB 53, the Transparency in Frontier Artificial Intelligence Act, took effect January 1, 2026 and applies to the largest frontier developers. It requires published transparency reports, catastrophic-risk assessments, critical-incident reporting, and whistleblower protections, with penalties up to $1 million per violation. It is not preempted; it remains enforceable, and California’s attorney general is actively opposing federal preemption.
Comply with the state laws actually in force rather than the executive order’s stated goal. Build to the strictest live regime, instrument for transparency-style obligations like incident reporting and model documentation, and treat federal preemption as a multi-year uncertainty. No state AI statute has been suspended by EO 14365, so the patchwork is still your compliance baseline through at least 2027.
Primary sources
- Federal Register: Ensuring a National Policy Framework for Artificial Intelligence (EO 14365) — Federal Register
- DOJ creates task force to challenge state AI regulations — CBS News
- Senate votes 99-1 to remove AI moratorium from Big Beautiful Bill — Reason
- State AI law moratorium omitted from 2026 defense bill — StateScoop
- Colorado Attorney General Delays Enforcement of Colorado AI Act — Troutman Pepper Locke
- Governor Newsom signs SB 53, advancing California’s AI industry — Office of the Governor of California
- Executive Order Tries to Thwart Onerous AI State Regulation — Crowell and Moring LLP
Last updated: May 31, 2026. Related: Governance.
