Federal legislation should establish a floor for AI enforcement, empowering states to go further.
Should the U.S. Congress preempt the growing patchwork of state artificial intelligence (AI) laws? The debate reached its first peak a year ago when, in July 2025, the U.S. Senate voted 99- 1 to strip a proposed 10-year moratorium on state AI law enforcement from President Donald J. Trump’s mega tax and spending law. The provision, championed by U.S. Senator Ted Cruz (R-Tex.), would have conditioned access to federal broadband and AI funding on states pausing enforcement of their own AI laws. Even a year later, the debate continues. The FRONTIER Act, which was introduced in late July 2026 as a rework of the earlier Great American AI Act, would preempt new state obligations on “frontier AI” developers in three defined subject areas: catastrophic-risk transparency, third-party auditing, and incident reporting. Its main sponsors, U.S. Representatives Jay Obernolte (R-Calif.) and Lori Trahan (D-Mass.), are reportedly pushing for a U.S. House of Representatives Energy & Commerce Committee markup when the House of Representatives returns for the regular annual session in September 2026.
On the one hand, a single national standard has real appeal: compliance clarity and no 50-state maze for AI innovators. And the current FRONTIER Act’s preemption is narrower than earlier drafts. It displaces state law only where a measure targets AI developers and falls within one of three covered subject areas, expressly preserving generally applicable laws, state regulation of deployers and users, child-safety rules, and state procurement. But even this targeted approach is ceiling preemption. Within the covered areas, states could not exceed federal requirements at all. Congress should instead set a federal floor that states may exceed within limits. Clear, congressionally enacted rules would provide greater certainty than volatile, executive-led decision-making while avoiding unnecessary federal overreach. And one condition is non-negotiable: wherever federal AI law imposes obligations on states or localities, it must fund them. Unfunded mandates would hollow out AI governance at the level where enforcement actually happens.
The case for preemption is legitimate. A single national standard prevents conflicting state rules from fragmenting a national industry, restrains outlier states from setting de facto national policy, and allows Congress to draw on federal expertise in areas such as national security. This is a recurring theme in federalism debates. Consider air travel. In 1958, Congress created the Federal Aviation Agency—now the Federal Aviation Administration—to establish a single, nationwide system of airspace regulation and improve aviation safety after a series of deadly military and civilian aircraft collisions. But aviation worked under full federal preemption in large part because the technology and its risks were well understood by 1958. With AI, we are in the opposite position: society does not yet fully understand the risks.
Rather than preempting states from creating their own standards, the federal government should set minimum standards that all states must meet. The case for preemption is even weaker when a federal prohibition on states’ regulation comes with no accompanying federal regulation, as in the summer of 2025. Subject to minimum standards, states would be free to go above the floor but within limits, such as prohibiting outright bans on lawful AI services. This system would allow for state-led initiatives while avoiding fragmentation.
Starting with a federal floor has decisive advantages. Chief among them, a floor—unlike blanket preemption—preserves states’ ability to keep experimenting above the baseline. States are often well-suited to serve as policy laboratories: essential testing grounds for regulatory approaches that Congress has historically been too gridlocked to address. At present, states are pursuing a range of approaches to AI regulation, creating opportunities to evaluate different policies under real-world conditions. California’s 2024 AI law imposes transparency requirements for AI training data. Colorado’s AI Act allows deployers to avoid the most severe liability by completing voluntary impact assessments. Illinois’s 2008 Biometric Information Privacy Act, which protects individuals’ biometric data, has been applied to data collected with AI. Each serves as a meaningful regulatory experiment. Preempting them before their outcomes are understood would destroy valuable opportunities to learn. And because policy changes are made locally, unintended consequences surface locally first, where they can be isolated and studied. This matters especially now, because society is still learning what AI harms even look like. A single federal ceiling would freeze the field prematurely, foreclosing that learning process. A federal floor, however, would keep it open.
In many areas of AI regulation, states and the federal government will cooperate, sharing the burden of regulation and enforcement. When this happens, though, one thing must be avoided: unfunded mandates. That is, when federal law requires states or localities to take on duties, whether audits, reporting, enforcement, or oversight, the money must come with it. Anyone who works alongside local governments sees this dynamic. In Minnesota, where I live and work, local governments are regularly charged with implementing state law without adequate funding: one recent state mandate imposed significant new documentation duties on county child-welfare staff. Counties warned that they could not comply, and implementation funding arrived only two years later, after concerted advocacy. But when funding does not come—or comes late—the results are often predictable: thin enforcement, overwhelmed local staff in many rural areas already facing declining working-age populations, and, in the end, mandates that exist on paper only. The same failure could occur with AI governance. If federal AI law treats the states as free labor, the enforcement of AI rules will fall on under-resourced state and local bodies. The federal Unfunded Mandates Reform Act has recognized this problem since 1995. But the Act only requires Congress to acknowledge costs, not cover them. If federal AI legislation is going to rely on states and localities for enforcement, the federal government should go further than the Unfunded Mandates Reform Act’s disclosure regime and attach funding. AI oversight is exactly the kind of technically demanding, capacity-hungry obligation that a mere point of order will not sufficiently support. If the United States wants state and local governments to play meaningful roles here, it must equip them with the expertise and resources to succeed.
In the end, Congress should not pass a blanket preemption on state AI regulation. States are policy laboratories where unintended harm first surfaces locally. A federal ceiling would freeze the field while society is still learning what AI’s harms even look like. Instead, the United States should set a federal floor that states may exceed within limits. AI governance will only be as strong as its weakest enforcement layer, and that layer is often going to be state and local. Wherever federal AI law imposes obligations on states or localities, it must fund them. A federal floor keeps the system adaptive, and funded mandates keep it enforceable. Good AI policy is not just about what a rule requires—it is about whether government can actually enforce it.
