The Trump administration released a four-page legislative framework on March 20 calling on Congress to establish a uniform national AI policy that would supersede state-level AI regulations it considers a barrier to U.S. competitiveness. The document is a blueprint for legislation, not a binding rule, but it represents the administration's most detailed statement yet on what federal AI governance should look like. It arrives after the administration's first attempt at preemption, a proposed 10-year moratorium inserted into the budget reconciliation package, was stripped from the bill by a 99-1 Senate vote last July.
The framework's central argument is that a federal standard must replace what it calls "a patchwork of conflicting state laws." The administration says states should not be permitted to regulate AI development and should not penalize AI developers for unlawful third-party conduct involving their models. At the same time, the document carves out exceptions: federal preemption would not extend to state child protection laws, consumer fraud statutes, zoning decisions for AI infrastructure, or rules governing states' own use of AI in procurement, law enforcement, and education.
For AI infrastructure specifically, the framework contains a provision that has received less attention than the preemption debate: a call for streamlined federal permitting to let data centers build or procure on-site and behind-the-meter power generation. The administration frames this as protecting residential ratepayers from bearing higher electricity costs driven by new data center construction. The practical effect, if enacted, would be to accelerate large-scale compute deployment by reducing the permitting friction that has slowed power interconnection agreements across multiple U.S. grid regions. White House Policy Director Michael Kratsios said the administration wants "one national AI framework, not a 50-state patchwork."
The framework's treatment of national security is brief but pointed. It calls for ensuring that national security agencies have sufficient technical capacity to understand frontier-model capabilities and risks, a framing that implicitly acknowledges the intelligence community's current gap in evaluating advanced AI systems internally rather than relying on outside assessments. The document also calls on Congress to augment the federal government's ability to combat AI-enabled impersonation scams, particularly those targeting seniors.
On copyright, the administration takes a pro-training position while leaving the legal question deliberately unresolved. The framework states that the administration believes training AI models on copyrighted material does not violate copyright law, but acknowledges that arguments to the contrary exist and says the courts should resolve the question. It explicitly asks Congress not to preemptively legislate in a way that would foreclose fair-use arguments for training data. The document simultaneously proposes a framework against unauthorized commercial use of AI-generated digital replicas of a person's voice or likeness.
Congressional prospects for a comprehensive federal AI bill remain unclear. The December 2025 executive order that preceded this framework directed the administration to prepare legislative recommendations and contemplated withholding federal support from states with "onerous" AI laws. But translating a four-page principles document into legislation capable of passing both chambers, particularly on a preemption provision that failed bipartisan scrutiny once already, represents a substantial political lift. States including Colorado, California, Utah, and Texas have already enacted private-sector AI rules, and those statutes remain in effect.
Sources: White House, Reuters, Politico
–
By the Control Plane Editorial Team