On July 1, 2025, the United States Senate voted 99 to 1 to strip a ten-year ban on state AI regulation out of the budget bill. That vote is the whole story of federal AI policy in miniature: everyone in Washington wants a national framework, and almost no one wants to be the one who takes the pen away from the states. What happened next is a case study in how an administration builds a framework anyway.
Step one was the December 11, 2025 executive order, "Ensuring a National Policy Framework for Artificial Intelligence." The policy sentence is one line:
It is 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.
Executive Order 14365 § 2 (2025)
The order can't preempt state law by itself, so it reaches for every lever short of a statute. It stands up an AI Litigation Task Force at the Justice Department whose only job is challenging state AI laws, on dormant Commerce Clause and preemption theories. It orders Commerce to publish a list of "onerous" state laws within ninety days, then makes states with those laws ineligible for certain broadband funds. It directs the FTC to issue a policy statement arguing that state laws requiring models to alter "truthful outputs" are preempted by the FTC Act's deception prohibition. That last theory is the cleverest move in the document: reframe a state safety requirement as a government mandate to deceive, and existing federal consumer protection law does the preempting.
Step two came on March 20, 2026, when the administration released the actual legislative blueprint, the "National Policy Framework for Artificial Intelligence." Four pages, seven pillars, no new federal AI agency: sector-by-sector oversight through existing regulators, industry-led standards, a statement that training on copyrighted material doesn't infringe (while deferring to the courts), and a ratepayer protection pledge so data centers pay their own power costs. And the pillar everything else scaffolding toward: Congress should preempt state AI laws that "impose undue burdens," while carving out traditional state police powers over children, fraud, and consumers, plus data center zoning and state procurement.
Read the carve-out closely, because it's where the real fight lives. The companion laws I wrote about in the last piece look a lot like consumer protection, which the framework says states keep. The frontier-model laws in California and New York look like AI development regulation, which the framework says states lose. Colorado's algorithmic discrimination act is the named villain of the executive order, and after a spring of amendments it was scaled back and pushed to January 2027. Whether the federal framework preempts anything will turn less on the word "preempt" than on where each state law falls on that map, and the Task Force's early docket will draw the first lines.
My read: comprehensive legislation doesn't pass this Congress, for the same reason the moratorium died 99 to 1. Preemption polls badly in both parties, the House and Senate want different bills, and the window closes at the midterms. What survives instead is the architecture the administration built without Congress: the Task Force, the funding conditions, the agency policy statements. The federal framework arrives as a set of levers, not a law.
For companies the practical answer hasn't changed: you comply with the strictest state you're in until a court says otherwise, and you watch the Task Force docket the way you used to watch the agencies. The patchwork everyone complains about is still the law of the land. What's new is that the federal government is now a litigant in it.