The White House released its National Policy Framework for Artificial Intelligence on March 20, 2026. The four-page document doesn’t change any law. It asks Congress to change them.
That distinction matters enormously for compliance teams.
The framework’s seven legislative pillars, child protection, AI infrastructure and small business support, intellectual property, censorship and free speech, enabling innovation, workforce preparation, and preemption of state AI laws, are recommendations, not requirements. Every single provision depends on Congress acting. Congress hasn’t.
The preemption provision is the one with teeth.
The framework calls on Congress to preempt state AI laws that “impose undue burdens,” while carving out state authority in areas including traditional police powers. According to Morrison Foerster’s analysis of the framework, this pillar sits alongside six others that together represent the administration’s full legislative agenda for AI. The preemption push didn’t appear in a vacuum. The framework builds on a December 2025 Executive Order that directed the administration to develop these Congressional recommendations, and that also deployed an AI Litigation Task Force with a specific mandate to challenge state AI laws the administration considers inconsistent with federal policy.
That task force is already operating. It’s a signal of intent even before Congress moves.
Why this is a political test, not a policy plan.
California has its own AI legislative agenda and shows no interest in yielding to federal preemption. Colorado already has an AI Act with compliance requirements that apply to high-risk systems. Other states have active or pending AI legislation. The administration’s framework would need Congress to pass legislation overriding these state laws, a heavy lift in any political environment, and a particularly uncertain one heading into a mid-term cycle.
The framework recommends “commercially reasonable, privacy-protective age-assurance requirements” for AI platforms likely accessed by minors, according to analyses by Holland & Knight and Crowell & Moring. The administration’s framework also states that training AI models on copyrighted material does not violate copyright law, though the document acknowledges courts are actively working through contrary arguments.
What compliance teams should do now.
Nothing in this framework is binding today. But the AI Litigation Task Force is real, and the administration’s stated position on state law preemption is real. Compliance teams at companies with multi-state operations face a genuine decision: build to state-specific standards now, build to whatever federal framework eventually emerges, or build to both and absorb the cost.
The safe answer, build to the stricter standard, isn’t always the practical answer. Watch for Congressional bill introductions that attempt to codify the framework’s preemption provision. That’s the moment the political test becomes a legal one.