White House framework asks Congress to preempt state AI laws (March 2026)

On 20 March 2026 the White House released its National Policy Framework for Artificial Intelligence, a short legislative recommendation prepared by science adviser Michael Kratsios and AI adviser David Sacks. The document fulfils a December 2025 executive order that directed the administration to draft a federal approach to AI and to challenge state laws seen as obstacles to the technology. Its central request is that Congress preempt state AI legislation and replace it with a light-touch national regime.

Under the proposal, states would keep authority over generally applicable laws, data-centre zoning and their own procurement, but would lose the ability to regulate AI development itself or to penalise developers for how third parties use their models. The framework rules out creating a new federal AI regulator. It pairs preemption with a set of targeted measures: child-safety requirements including age assurance (explicitly not age verification) and protections against exploitation; streamlined permitting for data centres alongside safeguards against residential electricity price increases; regulatory sandboxes allowing companies to seek exemptions from federal rules; deepfake protections with First Amendment carve-outs; and a voluntary copyright licensing framework.

Reaction split along familiar lines. Industry groups welcomed the text as necessary to avoid a patchwork of state rules, and House Republican leadership endorsed it immediately. Consumer and civil-society organisations warned that it would let companies ship harmful products without accountability. Whether the recommendations become law depends on Congress, where Senator Marsha Blackburn had circulated draft legislation but bipartisan support was uncertain, and where a similar moratorium on state AI laws had been stripped from a budget bill in 2025.

The framework is the latest turn in an escalating federal-versus-state contest. Colorado's AI Act, California's frontier-model transparency law and Texas's governance statute all took shape in 2024 and 2025, and several more states passed transparency measures in early 2026. The administration's December order had already instructed the Justice Department to litigate against state laws it deemed burdensome and to condition certain federal funds on state cooperation. The March framework moves the fight from executive action to a request for statute, which would be far harder to reverse in court.

For multinational organisations the practical effect is asymmetric: European obligations under the EU AI Act continue to firm up, while the US direction of travel is toward fewer binding development-side rules and more emphasis on child safety, content provenance and infrastructure.

What it means for leaders

• Keep state compliance programmes running. Preemption is a proposal, not law; Colorado, California and Texas obligations remain enforceable on their current schedules.
• Build one global control set. Anchoring on NIST AI RMF and ISO/IEC 42001 lets you satisfy EU AI Act requirements and any future US federal baseline without maintaining parallel programmes.
• Prepare for age-assurance and deepfake duties. These elements have broad support and are the most likely to survive in any federal bill; product teams handling minors or synthetic media should start design work.
• Watch liability allocation. The proposal shields developers from penalties for third-party misuse, which pushes responsibility toward deployers; review indemnities and acceptable-use clauses in AI contracts.
• Engage in the sandbox debate. Federal sandboxes could create regulated pathways for pilots in health, finance and energy; monitor how eligibility is defined.