In Brief
The Trump administration spent the past week floating, then walking back, the idea of an FDA-style federal review for new AI models. This move indicates the government is moving toward some control of frontier AI. The big question is what form will that take?
What Happened
The federal government took its first serious public step toward vetting frontier AI models, and then partially backed away within 72 hours. The New York Times reported on May 4 that the White House was considering an executive order to create a federal review process for new AI models before they are released. The plan would set up a working group of tech executives and government officials to design oversight procedures, with the framework being discussed with Anthropic, Google, and OpenAI.
On Wednesday, National Economic Council director Kevin Hassett confirmed on Fox Business that an executive order is being studied, and he likened it to FDA drug testing. The administration is also discussing tapping the intelligence community to assess models, partly so U.S. agencies can study new capabilities before Russia and China see them.
The industry pushed back immediately. Chief of Staff Susie Wiles posted that night saying the White House is "not in the business of picking winners and losers." By Friday night, Bloomberg reported the administration is preparing an order that would direct U.S. agencies to partner with AI companies on cyber defense, but stop short of requiring government approval for cutting-edge models.
SmarterX founder and CEO Paul Roetzer discusses the White House's moves and the reaction to them on Episode 214 of The Artificial Intelligence Show.
The Key Numbers
72 hours - Time from a FDA-style review proposal from the White House to it being walked back
3 - AI labs involved in the White House discussions: Anthropic, Google, and OpenAI
10% - the federal government's stake in Intel, the soft-nationalization precedent already set
25+ years - national security AI research captured in Annie Jacobsen's "The Pentagon's Brain"
Government Control Over AI Seems Imminent
The administration is testing messaging in public. The shape of the week looks less like a policy reversal and more like a controlled trial balloon. "They have to find some way to do this, but the administration definitely seems like it's trying to thread the needle, test different messaging points," says Roetzer. "It's like they put it out there, by Tuesday they're thinking about basically approving the models, and by Friday it's, no, no, no, we're still in draft form. I'm sure they got massive blowback from tech community."
A scientific review process is not what's on offer. The most important reason a true model-review is unlikely to work is political contamination of expert review. Roetzer points to a New York Times report this past week that the FDA blocked publication of several studies supporting the safety of widely used vaccines. "If we can't agree on what should be relatively objective, it is or is not effective against these conditions, I can't imagine a scenario in today's climate where these things would be unbiased and truly scientific," he says.
The Dean Ball framing matters. AI policy analyst Dean Ball published an essay this past week called Before Leviathan Wakes arguing the national security apparatus is going to assert itself over frontier AI no matter what, and the only viable path is to create private intermediary institutions that sit between the state and the labs. Ball quotes Tyler Cowen, who said "we thus want sustainable methods of perpetual interference that are actually somewhat useful from a safety perspective and give governments some control and feeling of control, but not too much control."
"Even without any formal nationalization efforts, the government already has tremendous ability to exert influence on these labs."
Paul Roetzer, founder and CEO at SmarterX, Episode 214 of The Artificial Intelligence Show
The government might just build its own. The other path is one not being talked about publicly. "My assumption is the government is already building their own lab," says Roetzer. "If they haven't, they will." He points to the federal government's 10% stake in Intel, the CIA's venture arm In-Q-Tel, and a 25-plus year track record of national security work on AI captured in Annie Jacobsen's "The Pentagon's Brain."
SmarterX Take
This is the first visible move in a much longer process. The administration tested FDA-style language, the industry pushed back hard, and the White House landed on a narrower cybersecurity directive. That cycle will repeat. Each round will push a little further into how frontier labs operate, what models they release, and what information goes to the intelligence community first.
For business leaders, the practical question is not whether government controls happen. It's how to operate within them. Expect more friction in model releases, more variation in what's available in different jurisdictions, and more cases where the same lab gets pulled in different directions by enterprise customers, regulators, and national security agencies at the same time. The labs that handle that tension cleanly are the ones to build on.
What to Watch
The final executive order. The narrower cybersecurity order Bloomberg flagged is the immediate next step, but a broader model-vetting framework is still being drafted in some form. Watch whether it emerges as a true intermediary structure of the kind Dean Ball describes or as something closer to direct federal review.
Quiet government capacity building. The bigger signal will be in budgets and contracts, not press releases. Watch for expanded compute procurement under the Defense Production Act, additional federal stakes in chip and infrastructure suppliers, and any indication that the intelligence community is building or licensing its own frontier models. That will tell you more about where the U.S. is actually heading than any single executive order.
Further Reading
New York Times: White House Considers Vetting AI Models Before Release → nytimes.com
Bloomberg: US Prepares AI Security Order That Omits Mandatory Model Tests → bloomberg.com
Politico: White House Mulls Tight New Controls on Advanced AI → politico.com
Politico: White House Plots New AI Oversight Push → politico.com
Dean Ball: Before Leviathan Wakes → x.com
Ars Technica: What Could Go Wrong With Trump's AI Safety Tests → arstechnica.com
Heard on The Artificial Intelligence Show, Episode 214
Paul Roetzer and Mike Kaput discuss the federal government's moves toward more control over AI, the industry's reaction to it and why the U.S. government might just build its own models. Listen →
Mike Kaput
Mike Kaput is the Chief Content Officer at SmarterX and a leading voice on the application of AI in business. He is the co-author of Marketing Artificial Intelligence and co-host of The Artificial Intelligence Show podcast.