White House Floats an FDA-Style Pre-Release Review for AI Models

The Trump administration is considering an executive order that would require new AI models to clear a federal review process before they go public, modelled loosely on the way the Food and Drug Administration vets new drugs. The idea was floated this month by National Economic Council director Kevin Hassett and confirmed in subsequent reporting by The Hill and Axios. According to those reports, a working group of tech executives and government officials would design the procedures, and models would be "released to the wild after they've been proven safe."

If the framing sounds familiar, that is because the broad strokes resemble what the Biden administration tried to set up in 2023 before being unwound by Trump's first AI executive orders in early 2025. The current pivot is being driven, according to Fortune, in part by concerns about Anthropic's Mythos model and broader worries about Chinese frontier capabilities. The about-face has not gone down easily inside the administration's own coalition.

An industry pushback that bit

Industry pushback against full FDA-style approval was immediate and well-organised. The American Enterprise Institute called the vetting proposal "bad policy", arguing that a centralised gatekeeper would slow innovation and entrench incumbents. Trade groups privately warned the White House that a pre-release approval regime would push frontier work to Europe, where the EU AI Act has already settled into a more predictable rhythm.

According to reporting from SmarterX, the White House has since walked back the FDA framing and is now landing on what one source described as a narrower cybersecurity directive. The current draft would compel labs to share certain pre-release safety evaluation results with a federal body, but stop short of requiring approval before deployment. That is closer to the voluntary commitments the major labs already make than to anything genuinely new.

The state-versus-federal collision

The proposal also lands in the middle of an unresolved fight between Washington and the states. Trump's December 11 executive order directed the Department of Justice to challenge state AI laws and tied broadband funding to alignment with a forthcoming federal standard. On March 20 the White House released its National Policy Framework for Artificial Intelligence, urging Congress to preempt state laws that "impose undue burdens." Colorado has just slimmed down its own AI bill in part to head off that fight, but the state's revised statute still takes effect June 30.

The political logic of an FDA-style federal review is partly about giving Washington a credible answer to states. If the federal government can argue that AI safety is now a national-security matter being handled at the model level, the case for letting cities and states write their own rules becomes harder to defend. That logic only works if there is something real to point to, which is why the draft order matters.

What labs are actually being asked to do

Reading between the lines of the various leaks, the practical ask appears modest: share dangerous-capability evaluations with the National Institute of Standards and Technology's AI Safety Institute before release, allow government red-teaming on a defined subset of models above a compute threshold, and accept a soft hold while the review runs. None of that is FDA approval. All of it is a meaningful change from the current voluntary baseline.

The interesting variable is the threshold. The frontier labs already do most of this for their largest models. The question is whether the order also catches mid-tier and open-source releases, which would bring in a much wider cast of companies and a much harder set of compliance arguments. Watch for the compute floor when the executive order lands.