President Trump signed an executive order on June 2, 2026 asking AI companies to voluntarily share their most advanced AI models with the government up to 30 days before providing access to other partners. If that sounds like a small ask—it isn’t. This is the Trump administration’s answer to a simple but terrifying question: what happens when AI models can exploit cybersecurity vulnerabilities faster than defenders can respond?

The order was driven by Anthropic’s Mythos model, which the company says can exploit cybersecurity vulnerabilities at an unprecedented pace. Mythos spooked both government and Wall Street. When an AI model can find and exploit security flaws faster than any human analyst, the existing paradigm of “patch Tuesday, move on” breaks down. The government wants to see what’s coming before it arrives.

What’s actually in the order:

The executive order requests that AI companies developing models with “advanced cyber capabilities” share them with the government 30 days before releasing them to other partners. It calls for a “cybersecurity clearinghouse” to help national security agencies improve their defenses. It also explicitly states that nothing in the order “shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement.” The emphasis is voluntary—at least for now.

The context matters: This comes after the Department of Commerce announced in May 2026 that major tech companies would share unreleased versions of their AI models with the government for national security evaluation. That announcement has since disappeared from the Commerce Department’s website. The executive order is attempting to institutionalize what was previously an ad hoc arrangement.

Industry responded positively. OpenAI called it “an important step forward.” Microsoft said it “welcome[d] this effort by the Administration.” Anthropic, whose model triggered the whole thing, was already giving select companies and governments early access to their most advanced models. This executive order essentially codifies and extends that practice.

But here’s the tension: The order is explicitly voluntary—and that might be its weakness. Companies can choose to comply or not. There’s no enforcement mechanism, no penalty for declining. What the government is banking on is that the largest AI companies will participate because they have incentives to: avoiding regulatory backlash, maintaining access to government contracts, and demonstrating good faith to a skeptical public.

The bigger story is what’s driving this: the pace of AI capability advancement is outrunning the government’s ability to secure critical infrastructure. When a single model can identify vulnerabilities faster than teams of security researchers, the old model of waiting for breaches and then patching breaks. The government needs to see what’s coming.

This isn’t heavy-handed AI regulation. It’s not the EU’s approach of hard rules and heavy fines. It’s something more subtle: an attempt to build situational awareness of AI capability advancement without stifling innovation. Whether that works depends entirely on whether companies actually participate.

The order explicitly says it’s not creating a mandatory pre-clearance requirement—for now. But the infrastructure is being built. If advanced AI models continue to pose novel national security risks, “voluntary” can become “mandatory” with a stroke of a pen.