White House AI plan sets up clash in Congress over federal vs. state rules

The White House on March 20 released a four-page National Policy Framework for Artificial Intelligence, the administration’s most detailed set of AI legislative recommendations to date. Prepared by OSTP Director Michael Kratsios and Special Advisor for AI and Crypto David Sacks, whose 130-day term expired days later, the document outlines 27 recommendations across seven policy areas and calls on Congress to replace a growing patchwork of state regulations with a federal standard.

The document builds on a December 2025 executive order that established a DOJ AI Litigation Task Force and instructed the Commerce Department to evaluate state AI laws within 90 days, a deadline the department has not yet publicly met.

The framework prioritizes child safety measures such as age-assurance requirements for AI platforms, parental controls, and safeguards against exploitation and self-harm. It calls for streamlined permitting of data centers and on-site power generation alongside protections to prevent higher electricity costs from shifting to residential ratepayers. On intellectual property, the administration maintains that training AI models on copyrighted material does not violate copyright law but urges Congress to let courts resolve the issue.

Why the White House rejects a new AI regulator

The framework explicitly opposes creating any new federal AI regulatory body, directing Congress instead to support sector-specific oversight through existing agencies and industry-led standards. Agencies such as the FDA, the FTC, and HHS would absorb expanded AI oversight without new statutory authority or dedicated funding.

On liability, the framework warns against “ambiguous standards about permissible content, or open-ended liability, that could give rise to excessive litigation,” and recommends that states not penalize AI developers for a third party’s unlawful conduct. Sacks, a venture capitalist and partner at Craft Ventures before joining the White House, had previewed this position in public remarks, aligning with Silicon Valley investors who argue that broad liability provisions would discourage AI investment.

The state preemption fight

More than 700 AI-related bills were introduced in state legislatures in 2025, and at least 18 states enacted substantive AI laws before Congress acted on any comparable measure. The framework recommends preempting state laws that impose “undue burdens” while preserving state authority over child protection, consumer fraud, and zoning. But it draws a firm line around AI development itself, calling it “an inherently interstate phenomenon with key foreign policy and national security implications” that states should not regulate.

Starting June 30, Colorado’s AI Act will require deployers of high-risk AI systems to exercise reasonable care against algorithmic discrimination in employment, housing, and healthcare decisions. The December executive order singled out Colorado’s law for requiring what the administration characterized as forced alteration of AI model outputs. Texas enacted its own Responsible AI Governance Act in June 2025, and California’s conversational AI disclosure laws took effect January 1.

Competing AI bills in Congress face a 60-vote hurdle

The Senate voted 99-1 last July to strip a proposed 10-year moratorium on state AI laws from the One Big Beautiful Bill. Sen. Ted Cruz (R-Texas), who chairs the Senate Commerce Committee, led the moratorium effort; Sen. Marsha Blackburn (R-Tenn.) introduced the amendment to kill it, calling the provision insufficient to protect “kids, creators, and conservatives.” A majority of Republican governors, led by Arkansas Gov. Sarah Huckabee Sanders, wrote to Congress opposing the measure.

Blackburn released her own 291-page discussion draft, the TRUMP AMERICA AI Act, two days before the White House published its framework. The bill incorporates the Kids Online Safety Act and the NO FAKES Act but diverges in notable ways, imposing a duty of care on AI developers and declaring that training models on copyrighted material does not constitute fair use, a position the White House leaves to courts.

On the same day, House Democrats introduced the GUARDRAILS Act to repeal the December executive order entirely, led by Rep. Don Beyer (D-Va.), with Sen. Brian Schatz (D-Hawaii) filing companion legislation. Any comprehensive federal AI bill requires 60 Senate votes during a midterm election year.

FDA, HHS, and FTC would take on expanded AI oversight

The FDA cleared a record 295 AI-enabled medical devices in 2025, bringing the total above 1,000, and is still finalizing lifecycle management guidance issued in draft form in January 2025. HHS released a department-wide AI strategy in December 2025 and opened a public comment period on accelerating AI adoption in clinical care.

The framework does not specify how these agencies should structure their expanded AI responsibilities, but its sector-specific language points toward a risk-based model in which each regulator applies existing domain expertise to AI applications within its jurisdiction.

Days after the framework’s release, Sacks confirmed his 130-day term as a special government employee had expired and moved to co-chair the President’s Council of Advisors on Science and Technology alongside Kratsios. At the Hill and Valley Forum on March 24, both officials identified child safety as the likeliest area for early bipartisan action.

The framework is not law, and its path through Congress remains uncertain. But it draws a firm line in the policy debate, positioning federal preemption and agency-led oversight as the administration’s preferred model as states continue to move ahead on their own.