Trump signed an executive order on Tuesday asking AI companies to share their most advanced models with the government up to 30 days before providing access to anyone else. The stated reason: cybersecurity risks and protecting critical infrastructure. The timing: right after Anthropic’s Mythos model spooked everyone with its ability to exploit cybersecurity vulnerabilities at unprecedented pace.

What’s Actually in the Order

The executive order requests that AI companies voluntarily share new models deemed to have advanced cyber capabilities with the government a month before releasing them to partners. A White House spokesperson called it a “common-sense approach of collaborating with industry to balance innovation and security.”

The key word is “voluntarily.” The order explicitly states that nothing in it “shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement.” This is a request, not a mandate. At least for now.

Major AI companies engaged with the White House on developing the order. That includes Anthropic, which the Pentagon blacklisted after a disagreement over guardrails in its models on classified military systems. OpenAI called it “an important step forward.” Microsoft President Brad Smith welcomed the effort. Everyone’s being very diplomatic about what is, at core, a significant expansion of government visibility into AI development.

Why Now?

The trigger appears to be Anthropic’s Mythos model, unveiled earlier this year. Mythos can exploit cybersecurity vulnerabilities at a pace that alarmed both government and Wall Street. When your AI model can essentially automate sophisticated hacking, the national security implications become immediate rather than theoretical.

The order had been expected nearly two weeks ago but was postponed. Trump himself said at the time he delayed it because he “didn’t like certain aspects” and thought it “gets in the way” of AI development. The version that finally emerged is shorter — 30 days instead of 90 — and explicitly voluntary. That’s a compromise.

What This Means

The practical impact right now is minimal. Companies aren’t required to share anything. But this establishes a precedent: the government expects to see advanced AI models before they’re released to the public. It’s a foot in the door.

The order also calls for national security agencies to improve their cybersecurity defence with a “cybersecurity clearinghouse.” That’s bureaucratic speak for a system to assess AI-powered threats. The government is building infrastructure to handle AI-augmented attacks.

And here’s what the order doesn’t say: anything about state-level AI regulation. Trump’s team has been trying to preempt states from regulating AI. Some states are forging ahead anyway. This executive order is federal, not preemption. It’s about the feds getting access, not blocking states.

The Bigger Picture

We’re watching the federal government establish a norm of advance review for advanced AI. Today’s voluntary 30-day window could become tomorrow’s mandatory 90-day review. The order explicitly preserves the option to create mandatory licensing “for future administration.”

If you’re building AI with significant capabilities, expect the government to want a look before the world does. This executive order is the opening move in what will likely be an ongoing negotiation over how much visibility the feds get into AI development. The question isn’t whether that happens — it’s how much and how fast.