As artificial intelligence evolves rapidly in the United States, policymakers are debating not only how the technology should be regulated, but also which level of government should have the authority over those rules.

Recent federal and state policy proposals have used preemption—or limiting and overriding the regulatory authority of lower levels of government—to govern the use, development, and regulation of AI platforms and data centers. Some advocates argue that preemption can be a comprehensive response to a patchwork of state and local laws governing AI, while others believe it could reduce opportunities for local voice and constituent protections, particularly in those states that oppose the federal AI framework.

As lawmakers continue developing legal frameworks for AI, they need to be clear about how and why they are using preemption to prevent community harm and to amplify the voices of those they represent.

Preemption can take different forms and produce different outcomes  

Not all preemption works the same way.  Floor preemption sets minimum standards that lower levels of government must follow while allowing them to adopt additional regulations. This approach aims to promote consistency while maintaining local flexibility. For AI, this could look like Congress or states requiring baseline safety testing, transparency measures, and risk assessments for AI systems, while still allowing states or localities to implement consumer protections, privacy requirements, or liability standards tailored to their communities.

By contrast, ceiling preemption limits or prohibits state and local action. Under this model, lower levels of government cannot enact regulations that go beyond, or sometimes even exist alongside, regulations set by federal or state governments. Some AI ceiling preemptions, including right to compute measures, may prevent states or localities from developing AI governance models tailored to local needs.  

Preemption has been used to both limit local regulation authority and protect it. As AI platforms, data centers, and uses in government continue evolving quickly, it is impossible for any one type of regulation to anticipate all its potential effects or consumer needs. In this moment, preemption may be functioning as more of a workaround to this rapid change rather than a governing solution itself.

Federal and state governments are already pursuing preemption on both sides of the aisle

At the federal level, an early draft of a bipartisan Congressional bill on AI safety and risk included many provisions to preempt states’ ability to impose rules on AI tools and developers. But after critiques from AI safety advocates and others (e.g., state lawmakers) about these preemptive ceiling provisions, lawmakers narrowed them substantially. The revised bill allows states to regulate AI uses, procurement, privacy, and consumer protection.  

States are having similar discussions. Conservative organizations such as the American Legislative Exchange Council (ALEC) have proposed model legislation aimed at identifying and reducing local barriers to AI development, deployment, and use. It has also proposed model legislation for preempting the right to compute, which would restrict government limits on computational resources except when demonstrated to be necessary and narrowly tailored to fulfill a “compelling government interest.” Meanwhile, progressive organizations, such as Local Solutions Support Center, are tracking preemption related to AI (among other topics), and consider preemption “abusive” if the law specifically blocks protections for families’ economic well-being.

Debates over data center development illustrate how preemption can produce different outcomes. Some states have sought to limit or prohibit building certain data center projects, while others have preempted local governments from imposing moratoriums or restrictions on new facilities and taxation (essentially limiting new limits). For example, states like West Virginia have proposed ceiling legislation that removes counties’ authority to review projects related to data center constructions within their jurisdictions. Meanwhile, others have proposed state moratorium bills to pause new data center projects, including New York’s latest one-year moratorium or Arizona’s three-year moratorium on the state’s data center sales tax exemption.

Preemption is not new, but its use for AI implicates more people

Limiting local regulatory power alone does not address the broader challenge of creating a comprehensive legal framework to inform AI decisionmaking. Evidence from other policy areas suggests preemption is not a blanket solution to complex governance challenges.

The preemption techniques being used to govern AI and data centers resemble preemption in housing policy. Though many localities face a serious housing crisis, local governments have limited ability to adopt policies that could address housing supply, price, and quality—such as source-of-income nondiscrimination, inclusionary zoning and other housing programs, and short term rentals and other rent regulations—because more states are preempting local ability to regulate. Meanwhile, state laws and regulations have not solved the affordable housing crisis, as members of many communities still face eviction, homelessness, and housing instability. 

Decisions about AI governance will not just shape the development of AI, but may also affect many other industries, so it is critical to examine how governance structures shape the economy, including market outcomes, competition, and innovation. Critics suggest blanket preemption can contribute to regulatory capture, in which regulations disproportionately serve the interests of dominant industry actors rather than promoting fair competition and efficient markets. For example, the State of New York preempted New York City’s plastic bag tax as a result of lobbying by special interest groups. 

Proponents of preemption argue that having a patchwork of regulations in different states is costly because it increases costs for entrepreneurs, who must comply with different laws in different states. But it is unclear whether preemption actually reduces these costs, particularly when it contributes to regulatory capture.

Preemption limits communities’ say in how to regulate and govern AI

Given preemption’s far-reaching effects on community well-being, it should be a narrow exception rather than a default policy response. As lawmakers develop legal frameworks on AI, they need to assess its effects—including whose voices go unheard when state and local regulations are limited—and explain how those choices still protect communities from AI harms, such as job losses or eroded data privacy.