Donald Trump signed an executive order this week asking — not requiring, asking — AI companies to share their most advanced models with the government up to 30 days before providing access to anyone else. The stated purpose: national security and cybersecurity. The unstated reason: probably Anthropic's Mythos model, and the genuine terror it has inspired in government and Wall Street.
Let's be precise about what this order actually does. It creates a voluntary framework. Companies are asked — not compelled — to share new models that have advanced cyber capabilities before general release. The order explicitly states that nothing in it should be construed as creating a mandatory licensing or preclearance requirement. It's a request dressed up as policy, backed by the implicit weight of White House attention.
Why Now?
The timing tells the story. Anthropic unveiled Mythos — a model it says can exploit cybersecurity vulnerabilities at an unprecedented pace — and suddenly the regulatory conversation in Washington shifted from "should we regulate AI?" to "how do we make sure nobody deploys a cyber weapon disguised as a chatbot?"
The Department of Commerce had announced last month that major tech companies would share unreleased AI models with the government for national security evaluation. That announcement has since disappeared from the Commerce Department's website. The executive order is an attempt to put something concrete in its place, or at least to be able to say the White House tried.
The earlier draft called for a 90-day review period. AI companies pushed back hard, arguing that 90 days in AI time is an eternity, and that such delays could hand competitive advantage to foreign rivals. The final order settled on 30 days — still meaningful, but far less onerous.
The China Shadow
Every conversation about AI regulation in the US eventually circles back to one question: what about China?
The executive order was postponed in late May, reportedly because Trump didn't like certain aspects he thought could be "blockers" to US AI development. The concern was explicitly framed — by Trump himself — as whether mandatory pre-release testing could let China pull ahead.
This is the tension at the heart of US AI policy right now. On one side: genuine security concerns about models that can find and exploit software vulnerabilities at superhuman speed. On the other: the geopolitical race for AI supremacy, where any regulatory friction is framed as a gift to Beijing.
The voluntary nature of this order is, in a sense, a compromise between those two pressures. It gives the appearance of doing something about AI security without imposing the kind of mandates that would make the industry scream and give China a relative advantage.
The Industry Response
Big Tech was... remarkably diplomatic. OpenAI called it "an important step forward." Microsoft said it "welcome[d] this effort." These are the words of companies that have been in the room, helped shape the policy, and gotten something they can work with.
Anthropic — the company that arguably triggered all of this with Mythos — is worth noting. The Pentagon blacklisted them after a disagreement over guardrails in their models on classified military systems. And yet they're part of the companies giving select governments and enterprises early access to their most advanced models. Their Mythos model is being released to all customers "in the coming weeks."
There's a pattern here. The government is trying to get ahead of risks that the market is moving too fast to address voluntarily. Whether this executive order accomplishes that — and whether "asking nicely" will be enough — remains to be seen.
The honest answer is probably not. But in the absence of a better alternative, this is where we are.
Top comments (0)