Executive Order 14365 (Ensuring a National Policy Framework for Artificial Intelligence)

Signed 11 December 2025. T1 — the White House's own published text. This is the order us-ai-policy had been describing for eight months as "the Dec 2025 EO" without a number, and its mechanisms are more specific than that page had recorded.

It is the deregulatory pole of US AI policy, and its target is not AI companies. It is the states.

Three levers, only one of them a courtroom

• An AI Litigation Task Force. The Attorney General must stand one up within 30 days to challenge state AI laws inconsistent with federal policy — on interstate-commerce or constitutional grounds.
• Money, which is the sharper instrument. Within 90 days the Secretary of Commerce must issue guidelines making states with "onerous AI laws" ineligible for remaining BEAD funds — the Broadband Equity, Access and Deployment programme, reported at $42 billion already allocated. Agencies must further review discretionary grants and may condition them on a state either avoiding conflicting AI laws or entering a binding non-enforcement agreement. A state can therefore keep its statute on the books and contract away its own enforcement, which is a quieter outcome than repeal and harder to see.
• A list. Commerce has 90 days to identify problematic state AI laws, singling out any that require models to "alter their truthful outputs" or that violate constitutional protections.

And a fourth, aimed at the only body that can actually preempt: the Special Advisor for AI and Crypto and the Assistant to the President for Science and Technology must jointly draft legislation establishing a uniform federal policy that preempts conflicting state law.

What it exempts, and what that reveals

Proposed federal legislation is directed not to preempt state laws on child safety, AI compute and data-centre infrastructure (general permitting excepted), and state procurement and use of AI. The carve-outs are the clearest statement of where the administration expects preemption to be politically unwinnable. Read them as a map rather than a footnote: protecting children, siting data centres, and a state's own purchasing.

It has not preempted anything

Preemption normally flows from an act of Congress. An executive order does not displace state law by itself, no federal AI statute exists, and commentary through mid-2026 reads state AI laws as largely unaffected in the near term. The order's own structure concedes this — it commissions legislation rather than claiming the effect.

So the US "market-driven / deregulatory" cell in this wiki's four-way comparison rests on an instruction to sue, a funding condition, and a bill that has not been written. That is a live push, not a settled position, and us-ai-policy now says so.

The pair it forms with eo-14409

Six months apart, and pointing opposite ways on purpose. EO 14365 works to remove constraints on AI conduct, by preempting the states. eo-14409 adds a security layer over frontier releases — voluntarily, as it turns out (crs-eo-14409-explained). The administration is not being inconsistent so much as separating the axes: conduct is deregulated by order, frontier release is handled by request and, in two cases, by ad-hoc action outside either order.

The federal-versus-state fight this order opens is the axis the EU, China and Israel cells of the comparison do not have — governance contested inside a jurisdiction rather than between them.

Related

us-ai-policy · eo-14409 · crs-eo-14409-explained · risk-based-regulation · ai-governance · eu-ai-act · synthesis