AI regulation

Before the election, the administration was building toward a single, light federal standard that would displace state AI laws. An executive order of 11 December 2025 directed the attorney general to create an AI Litigation Task Force to challenge state AI statutes as burdens on interstate commerce or as preempted by federal rules, and the Justice Department set it up in January (White House; DOJ). A White House legislative framework in March asked Congress to bar states from regulating AI development. The states kept legislating. California's SB 53 frontier-model transparency law took effect on 1 January 2026. Colorado replaced its 2024 AI Act with SB 26-189, scheduled to take effect on 1 January 2027, and xAI is suing to block the state's rules with the Justice Department intervening on its side (Colorado General Assembly; Reuters; Reuters).

The midterms close the legislative route to preemption for at least two years. Preemption needs a statute, and a Democratic House and Senate are unlikely to pass one that sweeps away California, New York and Colorado law, particularly after a campaign in which Democrats in Texas, Ohio, Wisconsin and Pennsylvania won by running on voters' unease about data centers (Politico). What Congress cannot do is withdraw the executive order or stop the Task Force from suing. Its lever is appropriations: a rider in the FY2027 Commerce-Justice-Science bill barring funds for suits against state AI laws is the most direct tool, and it would force Trump to choose between vetoing a funding bill and conceding the point. Democratic chairs can also subpoena AI companies, and Energy and Commerce Democrats want to examine AI-driven "surveillance pricing," an issue that draws in both Progressive Caucus chair Greg Casar and Republican Senator Josh Hawley; child-safety and deepfake bills also have Republican votes (Politico).

The executive can still press its case through litigation, procurement and grants. The Task Force can keep litigating, as it has in Colorado. Federal procurement rules can reward models and vendors that meet administration standards. Commerce can condition grants and broadband money on state restraint. A constrained president is more likely to use these tools than to abandon them, and the AI industry's political spending gives the White House allies who want preemption restored by 2029. AI super PAC networks including Leading the Future and Public First saw candidates they had backed in primaries win, while technology donors who funded Republicans, among them Musk, who spent about $224 million in Texas, saw most of those candidates lose (Politico).

For developers and deployers the practical result is the survival of the state-by-state regime that the industry spent two years trying to replace. California, New York and Colorado will keep their laws on the books, and Texas now has a Democratic governor-elect who campaigned against data-center deals. On current trajectory, divergent state rules persist through 2028 under a Congress more interested in investigating model safety, child protection and data use than in harmonizing them. The EU AI Act's general-purpose model obligations, already in force, remain the stricter benchmark for firms operating on both sides of the Atlantic. The link to data-center siting and electricity prices, which drove much of the voter anger, is developed in the cross-cutting section.