President Trump signed an executive order on June 2nd requesting that AI companies provide the government with access to their most advanced models — up to 30 days before public release — so that national security agencies can evaluate their capabilities. The order explicitly states that nothing in it shall be “construed to authorize the creation of a mandatory governmental licensing, preclearance or permitting requirement.”

In other words: this is voluntary.

Sure it is.

What the Order Actually Says

The order establishes a framework for the government — primarily the NSA and Cybersecurity and Infrastructure Security Agency — to develop a process for evaluating the “advanced cyber capabilities” of frontier AI models. Companies are asked to give the government early access, before wider release, so agencies can assess whether these models could be used to attack critical infrastructure, improve cyber offensive capabilities, or otherwise pose national security risks.

The order is framed around three realities the Trump administration has decided to accept openly: that frontier AI models now carry national security consequences, that the US government wants insight into those capabilities before they land in the wild, and that a formal pre-approval process would slow down American AI development in a way that would benefit China.

That third point explains the “voluntary” framing. The administration wants proximity to these models without the bureaucratic overhead of a formal licensing regime. In a sector where progress can happen in months, delay is itself a strategic cost.

The Elon Musk Problem

It’s worth noting what this order doesn’t do, because an earlier version apparently would have. According to reporting, the original draft required companies to submit models for testing 90 days before release — a more substantial gate. That version was killed after reported calls from tech executives including Elon Musk and Mark Zuckerberg.

This creates an obvious tension: the order is framed around protecting US AI capabilities from foreign adversaries, but it was also softened under pressure from exactly the kind of large American technology companies whose influence the national security apparatus presumably also wants to manage.

The simultaneous blacklisting of Anthropic from federal agencies — and Anthropic’s ongoing lawsuit challenging that designation — adds another layer of irony. The administration wants early access to Claude, but also believes Claude poses a supply chain risk to national security. These positions are not obviously compatible.

What “Voluntary” Actually Means

Here’s the thing about “voluntary” government requests in a market environment: they’re voluntary in law and optional in practice only if you don’t care about consequences.

If you’re a startup AI company building a cybersecurity product and the Department of Homeland Security is evaluating which vendors to trust for federal contracts, your willingness to participate in the government’s voluntary model evaluation program is not going to feel optional. If you’re preparing for an IPO — as Anthropic, OpenAI, and SpaceX all reportedly are — the question of whether you’ve been through government security evaluation is not going to feel like a suggestion.

“Voluntary” in this context means: the government gets what it wants without having to go through the legislative process that would make it mandatory. Companies that cooperate get access to procurement relationships, defense networks, and the implicit endorsement of being treated as a trusted partner. Companies that don’t will eventually notice the difference.

The Geopolitical Frame

What the order makes explicit is the degree to which AI has been moved into the same policy category as semiconductors, telecommunications, and critical minerals: technologies too important to be left to open market competition alone, and too strategically sensitive to trust entirely to foreign players.

The Diplomat’s analysis of the order frames it as moving toward a “two-track system” — US and allied models entering a trusted ecosystem built around government evaluation and security cooperation, while Chinese models face growing scrutiny over data flows, political alignment, and content control. Whether that system holds together is a different question. The current administration has shown, repeatedly, that it can be lobbied. The question is whether the next administration — or the one after that — maintains the same framework.

What seems permanent is the underlying logic: advanced AI is a national security asset, and the US government intends to treat it as one.

Sources: UPI, The Diplomat