Sources: the Trump administration is discussing an EO to create an AI working group to examine AI oversight procedures, including vetting models before release
The Trump administration, which took a noninterventionist approach to artificial intelligence, is now discussing imposing oversight …Forums:Hacker NewsForums:Hacker News:White House Considers Vetting A.I. Models Before They Are Released
New York Times
Context & Ripple Effects
This is an early marker of a shift from a stated noninterventionist posture toward federal processes for managing advanced-model risk. Related coverage shows the discussion broadening from a working group to possible voluntary pre-release access, release-timeline standards, and eventually consideration of an independent safety regulator.
The important unresolved question is institutional design: administration voices have rejected a centralized “FDA for AI,” while later reports point to oversight mechanisms that could still give government a standing role in model-release decisions.
First-order effects
The immediate effect is procedural: the White House would have a vehicle to assess oversight options and coordinate agencies around model vetting, rather than imposing a confirmed release rule.
Frontier-model developers face added policy uncertainty around what evidence, access, or review might be sought before public deployment if the effort advances.
Second-order effects
A federal review process would push AI companies to engage earlier with agencies on safety and release planning; the draft voluntary-framework and standards discussions in related coverage indicate that voluntary access may be the initial route.
Competing governance proposals—agency-led coordination, voluntary commitments, or an independent regulator—would determine whether compliance remains flexible or becomes a distinct pre-release gate for developers.
Third-order effects
If these steps consolidate, frontier-model releases could become an ongoing government-access and assurance process, even without the centralized regulator administration officials have publicly resisted.
The policy direction would shift AI competition toward firms that can support structured safety review and government engagement; whether that becomes binding depends on the eventual authority and design of the oversight body.
The trend: This is part of the move from broad AI-policy principles toward operational federal oversight of access to, testing of, and release decisions for advanced models.
NYT: The admin is discussing an executive order to create an A.I. working group that would bring together tech executives and government officials to examine potential oversight procedures, according to U.S. officials.... Among the potential plans is a formal government review
On Trump's first day back in office, he eliminated a Biden-era mandate for AI firms to share their security test results with the govt, calling it unduly burdensome. — Now, after Mythos, the WH is considering much more onerous AI regulation. — Remarkable about-face. — www.n…
I hope @pmarca is able to safely flee the country now in case the White House does something he said “would impose tyranny far beyond anything even imagined by the Communists and Fascists of the 20th Century” [image]
Today's Hyperdimensional anticipates the admin taking a much stronger role in AI regulation and proposes a light-touch way forward. Overreaction—*regulation much more onerous and intrusive than what Ben Buchanan and I suggested in NYT today*—is a very real possibility. [image]
Fears of AI-enabled cyberattacks are driving a rethink within the Trump administration of how to handle the release of new models and whether some rules or regulations may in fact be necessary. https://www.nytimes.com/...
Waiting to see more details, but any sort of preemptive, pre-release “vetting” of AI models by the White House could be tantamount to a de facto licensing regime, which should certainly not be done via executive orders. The Trump administration's first order of business on AI [im…
I'm utterly fascinated by the modern art of journalistic simulacrum — at no point do this piece's four authors consider wandering anywhere close to the idea of corruption, or that Trump says a lot of things that are utterly meaningless, or that our regulators no longer function…
This is a sorely needed regulatory pivot, with substantial geopolitical implications. If the US government vets AI models pre-release, and presumably requires companies to include certain safeguards, it also needs a global plan to preserve their security post-release. It is
Plenty on the tech right condemned Biden's requirements for pre-model-release safety disclosures to that administration (rightly so IMO). Will they also condemn the even more constitutionally dubious idea of pre-release model review by this administration?
My light-touch proposal is to use DPA Title VII to compel dual-use AI model developers to disclose their largest training runs (say, >=10^26 FLOPs) and the results of any internal testing. No onerous obligations and would automatically exempt little tech. Worth considering! 😅
“Vetting” is just another word for “licensing.” When governments licensed expressive technologies in the past, they used their power to establish orthodoxies. It took civil libertarians centuries to fight off prior restraints on speech. That work continues . . .
“White House Considers Vetting A.I. Models Before They are Released” is currently the top story on the NYT website. Another striking indication that we have reached “the midgame” or “AI policy without training wheels” [image]
This NYT piece gets something quite importantly wrong, btw — the UK has no rules on predeployment evals or safety standards, and everything I've heard is that we're a long way off from having them. [image]
Also... kinda seems worth noting that half the reason so many Silicon Valley VCs lined up behind Trump was because a pretty toothless AI policy plan from Biden that basically told companies “be more thoughtful.” If the Biden admin had talked about pre-vetting models, it would ha…