Trump’s AI executive order preempts state laws, shifts power to Washington

The most consequential AI policy move of the Trump administration may have arrived in a single sentence buried in a December 2025 executive order. The order, titled “Ensuring a National Policy Framework for Artificial Intelligence,” establishes a national policy to promote U.S. leadership in AI by preempting conflicting state laws, according to the Eversheds Sutherland tracker of Trump administration executive actions. The preemption clause is the headline. It shifts the center of gravity for AI regulation from Sacramento, Albany, and Austin to Washington, D.C.

The order sits alongside a cluster of AI-related directives signed in late 2025 and early 2026. The administration issued orders on “Promoting the Export of the American AI Technology Stack” (July 2025), “Accelerating Federal Permitting of Data Center Infrastructure” (July 2025), “Preventing Woke AI in the Federal Government” (July 2025), and “Launching the Genesis Mission” (November 2025), which tasks the Department of Energy and its national laboratories with using AI and supercomputing to accelerate scientific discovery. The December preemption order is the capstone: it tells every state that their patchwork of AI bills, from Colorado’s AI consumer protection law to California’s proposed safety testing mandates, now sits on uncertain legal ground.

The preemption strategy is not subtle. It asserts that AI is a matter of national economic competitiveness and national security, and that 50 different state regulatory regimes would fracture the market, slow deployment, and hand advantage to China. The order likely relies on the Commerce Clause and the doctrine of field preemption, arguing that the federal government occupies the entire field of AI regulation. The legal theory is aggressive. It invites immediate court challenges. Colorado, which passed the nation’s first comprehensive AI law in 2024, is an obvious plaintiff. So is California, which has debated multiple AI safety bills in successive sessions.

For AI builders, the order is a double-edged sword. The upside is clear: a single set of rules, not 50. Startups and labs that build AI products for a national market no longer need to track a dozen different state disclosure requirements, risk-assessment mandates, and model-testing obligations. The compliance cost of Colorado’s law alone, which requires developers of high-risk AI systems to conduct impact assessments and disclose those assessments to the state, runs into the hundreds of thousands of dollars per product line. Multiply that across states, and the savings from preemption are real.

The downside is that the federal framework may be weaker than the strongest state laws. Colorado’s law, imperfect as it is, requires developers to identify and mitigate algorithmic discrimination. California’s proposed bills would have required frontier labs to test models for catastrophic risks before training. The Trump administration’s approach, as signaled by the “Preventing Woke AI” order, is hostile to any regulation that touches on bias, fairness, or equity. The preemption order does not replace state laws with a robust federal regime. It replaces them with a policy of non-regulation, or regulation that is limited to national security and export control.

The timing matters. The order was signed on December 11, 2025, four days before the extension of the TikTok enforcement delay, and in the same month as orders on 6G wireless and space superiority. It is part of a broader push to consolidate federal authority over technology. The “Accelerating Federal Permitting of Data Center Infrastructure” order, from July 2025, streamlines permitting on federal lands and removes environmental and DEI-related regulatory hurdles. The “Promoting the Export of the American AI Technology Stack” order establishes the American AI Exports Program to push full-stack U.S. AI technologies overseas. The December preemption order ties the bundle together: one federal voice on AI, one set of rules for building it, one approach to shipping it abroad.

The legal challenge tracker that Eversheds Sutherland maintains flags executive orders currently or previously challenged in court. The “Adjusting Imports of Semiconductors” order, which imposed a 25% tariff on certain advanced chips in January 2025, carries an asterisk. The December preemption order does not yet carry one. That will change. The question is which court hears the first challenge and what standard of review it applies. The Supreme Court’s recent decisions on the major questions doctrine, which requires Congress to speak clearly when authorizing agencies to act on issues of vast economic and political significance, could give states a powerful argument: preempting all state AI laws is a major question, and the executive order lacks clear congressional authorization.

The practical effect for AI builders depends on where they sit. A startup training a foundation model in San Francisco now faces a single federal regulator instead of the California legislature. That is a win for speed. A startup building an AI hiring tool in Denver no longer needs to comply with Colorado’s impact assessment requirement. That is a win for cost. But a startup that wanted a clear federal standard for algorithmic fairness, or a requirement to test models for catastrophic risks, gets neither. The preemption order does not create a federal floor. It removes the state ceilings.

The order also intersects with the Genesis Mission, which channels federal compute and data through the DOE labs. A national policy framework that preempts state law gives the federal government more control over where and how AI is built. The labs get clearer marching orders. Industry gets a single point of contact. But the labs are not neutral ground. They are subject to the same political winds that produced the “Preventing Woke AI” order, which bans federal agencies from procuring LLMs that incorporate DEI frameworks. The same logic could extend to the compute resources and data sets the labs provide.

The most immediate signal for the industry is the end of state-level experimentation. For the last four years, states acted as laboratories of democracy for AI regulation. Colorado tested impact assessments. California debated compute thresholds. New York considered biometric privacy rules. The December order says the experiment is over. The federal government will set the terms, and the terms are designed to maximize deployment speed and export volume, not safety or equity.

That is the bet. It is a coherent bet, grounded in a real concern about U.S. competitiveness. It is also a bet that the public, or the courts, will accept a federal framework that preempts stronger state protections without replacing them. The first lawsuit will test that bet. The second will test whether the order survives a change in administration. A future president could rescind the order on day one, but the infrastructure of federal preemption, once established, is hard to undo. States that dismantled their AI regulatory offices will not rebuild them overnight.

The December order is not the end of the story. It is the beginning of the legal fight that will define AI governance for the next decade.