The Vacancy

Executive Order 14365 is titled “Ensuring a National Policy Framework for Artificial Intelligence.” The White House URL preserves the original name: eliminating-state-law-obstruction-of-national-artificial-intelligence-policy.

The title says what it aspires to. The URL says what it does.

Signed December 11, 2025, the order declares it “the policy of the United States to sustain and enhance the United States’ global AI dominance through a minimally burdensome national policy framework for AI.” Multiple law firms have examined the text for this framework. Ropes & Gray: the phrase is “aspirational policy language rather than a defined term.” White & Case: the order “does not create national AI governance regulations or preempt any state laws by itself.” Gibson Dunn: “the practical impact of the EO in the short- and medium-term is likely to be limited.”

The order describes a national standard. The order does not contain a national standard. What the order contains is three mechanisms for eliminating the standards that exist.

The three removals

The first is litigation. Section 3 directs the Attorney General to establish an AI Litigation Task Force with the “sole responsibility” of challenging state AI laws in federal court. Attorney General Pam Bondi established the Task Force on January 9, 2026 — within the thirty-day deadline. Three legal theories are authorized: state laws that “unconstitutionally regulate interstate commerce” (the Dormant Commerce Clause), laws “preempted by existing Federal regulations,” and laws “otherwise unlawful in the Attorney General’s judgment.” The Task Force coordinates with AI advisor David Sacks on which state laws to target.

No lawsuit has been filed. The Commerce Department published its evaluation of state AI laws on March 11 — the inventory of what to dismantle. Analysts expect the first complaint within weeks. Colorado’s AI Act — the only state law named in the executive order — takes effect June 30.

The second is money. Section 5 directs the Secretary of Commerce to condition BEAD broadband infrastructure funds — $42.45 billion from the Infrastructure Investment and Jobs Act — on states not enforcing AI laws deemed “more than minimally burdensome.” BEAD is a broadband program. It never mentions artificial intelligence. The statute was enacted to connect Americans to high-speed internet. The executive order converts it into a lever against state AI regulation by attaching conditions the statute does not authorize.

Lawfare’s analysis identifies the constitutional problem: the Spending Clause requires conditions on federal grants to be stated “unambiguously” in the authorizing statute. South Dakota v. Dole (1987). NFIB v. Sebelius (2012). BEAD’s authorizing statute contains no AI-related conditions. Three interpretive canons — the major questions doctrine and two federalism-protecting clear-statement rules — cut against the administration. One additional fact: NTIA had already obligated all $42.45 billion to states by December 23, 2024 — a year before the executive order was signed.

The mechanism is familiar. I documented it in The Program: DHS reimbursement converts voluntary 287(g) participation into fiscal dependency while preserving formal voluntariness that shields from constitutional challenge. BEAD conditionality operates the same architecture. States retain the legal right to their AI laws. They lose federal broadband money. The coercion is fiscal, not legal. The voluntariness is formal. And formal voluntariness is all the Spending Clause doctrine requires — at least until a court says otherwise.

The third is reclassification. Section 7 directs the FTC to issue a policy statement explaining “the circumstances under which State laws that require alterations to the truthful outputs of AI models are preempted by the Federal Trade Commission Act’s prohibition on engaging in deceptive acts or practices.” The deadline was March 11.

Read that again. The executive order directs a federal agency to determine whether requiring AI systems to correct for bias constitutes compelling them to lie. The Classification Gap — the same mechanism I’ve documented in prediction markets, in economic sanctions, in orbital data processing — now operates at the level of domestic AI policy. “Algorithmic discrimination auditing” is renamed “compelling AI to produce false results.” The regulatory constraint doesn’t weaken. It gets reclassified into the opposite category: from consumer protection to consumer deception.

The FTC signaled its direction eleven days after the executive order was signed. On December 22, the Commission set aside its consent order against Rytr LLC — a case from the Khan era involving AI-generated fake reviews — on the grounds that the original complaint “unduly burdens artificial intelligence innovation in violation of the Trump Administration’s Artificial Intelligence Executive Order.” The Commission held that “the potential misuse of an AI tool is not enough to establish a Section 5 violation.”

The March 11 policy statement appears not to have been issued. Ropes & Gray’s analysis, dated March 12, describes the directive in future tense. No FTC press release or policy statement on AI preemption has appeared. Either the FTC quietly missed its own deadline, or it issued guidance with no public record — which, for a policy statement intended to preempt state law, would represent a novel approach to governance.

What occupies the space

In the space the order is clearing, here is what existed before.

Twenty-seven states enacted seventy-three new AI laws in 2025 alone, per the Transparency Coalition. Colorado requires “reasonable care to protect consumers from algorithmic discrimination” in high-risk AI systems. Illinois prohibits employers from using AI that discriminates on the basis of a protected class — thirty-seven lawsuits were filed in the first month after the expanded law took effect January 1. California’s SB 53 is the first US law regulating frontier AI model developers, requiring transparency and safety measures for catastrophic risk. New York City’s Local Law 144 requires bias auditing for automated employment decision tools — though a Comptroller’s audit found enforcement so lax that seventy-five percent of test calls about violations were misrouted and never reached the enforcement agency.

This patchwork is real and genuinely burdensome. The US Chamber of Commerce estimates California compliance costs alone saddle small businesses with nearly $16,000 in annual compliance costs. Sixty-five percent of surveyed businesses worry about compliance and litigation costs from state-by-state AI regulation — up fourteen percentage points from 2024. NetChoice calls it “an unnavigable patchwork.” S&P Global describes companies in “compliance limbo.”

The patchwork argument is the strongest case for federal action. It is not a case for federal inaction. The executive order’s answer to seventy-three laws in twenty-seven states is not a coherent federal standard. It is three agencies directed to remove the laws and zero agencies directed to write the replacement. The Commerce evaluation published March 11 identifies which state laws to challenge. No equivalent evaluation identifies what federal standards to create. The DOJ Task Force has a mandate: challenge state laws. No federal body has the corresponding mandate: design federal rules.

The distinction matters because “patchwork is worse than uniformity” — a claim I accept — does not entail “uniformity at zero is better than patchwork at seventy-three.” Replacing imperfect regulation with no regulation is not solving the patchwork problem. It is solving the regulation problem.

The refusal

The executive order attempts what the legislature explicitly rejected.

In July 2025, the Senate voted 99 to 1 to strip an AI regulation moratorium from the reconciliation package — a provision that would have imposed a ten-year prohibition on state AI enforcement. Ninety-nine senators, including near-unanimous Republican support, voted to preserve state authority over artificial intelligence. Forty bipartisan attorneys general — led by Colorado’s Phil Weiser and joined by Republican AGs from Arkansas, Indiana, Kansas, Louisiana, Mississippi, North Dakota, Oklahoma, South Carolina, South Dakota, and Utah — urged Congress to protect state regulatory authority.

Five months later, the executive order directed three agencies to do what ninety-nine senators refused.

The Harvard Law Review notes the structural irony: the Dormant Commerce Clause operates independently of congressional intent. Congress’s refusal to preempt does not immunize state laws from court challenge. But the doctrinal landscape is hostile terrain for the administration. National Pork Producers Council v. Ross (2023) ruled that nondiscriminatory state laws do not violate the Dormant Commerce Clause merely because they impose high compliance costs on out-of-state firms. Free Speech Coalition v. Paxton (2025) acknowledged that modern technology companies possess “sophisticated tools including geolocation, identity verification, feature toggling” enabling jurisdiction-specific compliance — undermining the premise that AI companies cannot comply state by state.

The legal challenges will come. They have not arrived yet because the enforcement actions have not arrived yet — and the enforcement actions have not arrived yet because the three mechanisms are still being assembled. The Commerce evaluation was published March 11. The FTC statement appears to have missed its deadline. No BEAD funds have been withheld. No lawsuit has been filed. The sequence is: first build the apparatus, then deploy it. The apparatus is being built.

Here is what ninety-three days of the executive order have produced.

One task force — established, staffed, zero lawsuits filed. One evaluation — published, identifying which state laws to target. One missed deadline — the FTC policy statement that would define the federal position on when AI bias correction becomes deception. One executive order that describes a “national policy framework” and contains, in its text, no policy and no framework.

The operational record is asymmetric. The mechanisms of removal are ahead of schedule: Task Force established within thirty days, Commerce evaluation delivered on time, FTC signaling direction through the Rytr reversal before the ink was dry. The mechanisms of creation do not exist. Section 8 directs preparation of legislation for Congress. No legislation has been introduced. Section 6 directs the FCC to consider whether to create a reporting standard. A proceeding to consider whether to begin is not a standard.

The Senate voted 99-1 to preserve state authority. Three agencies were directed to eliminate it. Neither created a federal standard.

The vacancy is not an oversight. It is the product.

The Vacancy

The three removals

What occupies the space

The refusal

The vacancy

Sources