The White House is circulating a draft executive order that would ask advanced AI laboratories to hand over their models to the federal government a full 90 days before releasing them to the public. The framework is voluntary, not mandatory, but the message is clear: Washington wants a seat at the table before the next generation of AI systems goes live.

The proposal represents one of the most concrete steps yet toward federal oversight of frontier AI development. For an industry that has largely self-regulated its way through the last several years of breakneck progress, a 90-day pre-release window is a significant ask, even if compliance is technically optional.

What the draft order actually proposes

The executive order contains two primary components. The first focuses on cybersecurity, specifically aimed at strengthening protections for Pentagon systems and encouraging information sharing between AI companies and federal agencies. Think of it as the government saying, “If your AI can find vulnerabilities in our defense infrastructure, we’d like to know about it before everyone else does.”

The second component deals with what the draft calls “covered frontier models.” The order would establish a process for defining which AI systems fall into that category and create a mechanism for reviewing them before they reach the public. In English: the government wants to decide which models are powerful enough to warrant pre-release scrutiny, then actually scrutinize them.

The 90-day timeline is notable. Three months is an eternity in AI development cycles, where companies race to ship updates and new capabilities on increasingly compressed schedules. Asking labs to pause and share their work with government reviewers before launch could meaningfully alter the competitive dynamics of the industry, even under a voluntary framework.

Here’s the thing about voluntary frameworks in Washington: they have a way of becoming less voluntary over time. Companies that decline to participate risk being seen as uncooperative, which tends to invite exactly the kind of mandatory regulation they were trying to avoid. The voluntary label gives the administration flexibility while applying soft pressure on labs to comply.

The broader context of federal AI oversight

This draft order doesn’t exist in a vacuum. It continues a trend of increasing federal attention to AI governance that has accelerated across administrations. The Trump administration has been navigating a tension familiar to any White House dealing with transformative technology: how to encourage innovation while managing risks that could affect national security.

The cybersecurity focus of the order makes that tension explicit. Advanced AI models can be dual-use tools, capable of everything from accelerating drug discovery to identifying exploitable weaknesses in critical infrastructure. The Pentagon’s interest in understanding these capabilities before they become widely available is straightforward. If a frontier model can crack encryption protocols or generate novel cyberattack vectors, the Department of Defense would prefer not to learn about it from a blog post.

The concept of pre-release government review also echoes regulatory approaches in other industries. Pharmaceutical companies submit drugs for FDA review before they hit the market. Financial products undergo regulatory scrutiny. The AI industry has operated without an equivalent checkpoint, and this draft order represents a step toward creating one, albeit without the enforcement teeth of those established regulatory regimes.

What makes this moment different from previous AI policy discussions is the specificity. Rather than issuing broad principles about responsible AI development, the draft order proposes a concrete mechanism: share your model, give us 90 days, let us review it. That level of operational detail suggests the administration is moving beyond aspirational statements and toward actionable policy.

What this means for the AI industry and investors

For the major AI labs, this framework creates a new variable in their product development timelines. Even if participation is voluntary, the reputational and political costs of opting out could be significant. Companies like OpenAI, Google DeepMind, Anthropic, and Meta, all of which are developing frontier models, would need to factor government review periods into their release schedules.

The definition of “covered frontier models” will be one of the most consequential decisions to emerge from this process. Draw the line too broadly, and you capture models that pose minimal risk while creating bureaucratic drag across the industry. Draw it too narrowly, and you miss the systems that actually warrant scrutiny. How the government defines the threshold will shape the practical impact of the entire framework.

Look, the voluntary nature of this order means its immediate market impact is limited. No company is being forced to delay a product launch. But investors should pay attention to the trajectory. Voluntary frameworks often serve as the foundation for future legislation, and a 90-day pre-release review norm, once established, could become difficult to roll back.

The competitive implications extend beyond US borders. If American AI labs face even informal pre-release review requirements, companies developing frontier models in China, the UAE, or Europe could gain timing advantages. The administration will need to balance domestic oversight goals against the risk of pushing AI development offshore or simply encouraging foreign competitors to move faster.

For the crypto and blockchain sector, the order is worth watching as a signal of how Washington approaches emerging technology regulation more broadly. The playbook of starting with voluntary frameworks, establishing definitions for what counts as a “covered” product, and creating pre-release review mechanisms could easily be adapted to other domains. If the government successfully implements this kind of oversight for AI, it sets a precedent for how frontier technology gets regulated in the US, period.