Trump signs order on voluntary vetting of AI models

Is the government now looking over the shoulder of every new AI model before it reaches you? Not quite, but it just moved closer than ever.

Donald Trump signed an executive order on June 2 allowing US companies to voluntarily share their most powerful AI models with the government for safety review. The number that matters is the window: the government gets access up to 30 days before a model is released to the public. An earlier draft set that window at 90 days, so this is a noticeably lighter touch toward firms worried about being slowed down right before launch.

What does the order actually change?

The most important part is what the order deliberately avoids. There is no mandatory licensing. There is no prior approval a company must obtain before shipping a model. The government cannot say you may not release this, it can only look earlier and flag concerns.

In practice, the order sets up a few things:

• A voluntary early look at the most powerful models, up to 30 days before launch.
• The creation of an AI cyber clearinghouse, the venue where that review happens.
• Oversight handed to the NSA director, which signals the focus is on cybersecurity rather than content control.
• A task for agencies to develop benchmarks for assessing the cyber capabilities of AI models, so there is something concrete to measure.

The White House drew a clear line away from heavy-handed regulation.

We are not conducting oversight of every new model, because that level of government interference would have a chilling effect on free speech and innovation.

That is a shift from the earlier hands-off stance, but a carefully dosed one. Rather than leaving the market entirely to itself or imposing hard permits, the order picks the middle: the government may peek, but it may not block.

The road to the signature is telling too. Trump refused to sign an earlier version on May 21, the one with the 90-day window and a wider reach. He signed the revised, softer version only on June 2. The gap between those two drafts is the whole story: how far the state may slow an industry that describes itself as central to national security.

Why should you, the user, even care?

Because this sets the template for how the strongest AI tools will be supervised, and you already use those tools every day. If the review catches a serious cyber risk before launch, the model you get is safer. If the review turns into a narrow political lever, you get slower development. The line between the two outcomes is thin.

For now the industry is cautiously positive. OpenAI backed the idea that the framework should emerge through institutions rather than a one-sided government decree.

We believe effective safety frameworks should emerge through democratic institutions.

That was Chris Lehane of OpenAI, and the sentence captures the compromise tone of the whole order. No one got everything they asked for. The government did not get hard control, the companies did not get full freedom, and the user sits somewhere in between, waiting to see how voluntary vetting plays out in practice.

Realistically, the first few models that pass through the new cyber clearinghouse will show whether this is a serious review or a formality on paper. Until then, it is all a promise.

So here is the question for you: would you rather use an AI the government looked at before launch, or one that arrived with no review at all?