An AI policy draft appears to be softening
A draft executive order being developed by the Trump administration would reportedly make government review of frontier AI models voluntary rather than mandatory, according to leaked details described in recent media reporting. If that characterization proves accurate, it would mark a significant shift from earlier expectations that the order could create a more forceful federal screening regime for advanced models.
The reported change matters because frontier-model oversight has emerged as one of the central fault lines in U.S. AI policy. A system that requires developers to submit powerful new models for government vetting would represent a much stronger regulatory intervention than one that merely invites them to do so. A voluntary framework, by contrast, depends heavily on company incentives and reputational considerations rather than legal obligation.
At this stage, the key limitation is clear: the order itself has not been published, and the available details are attributed to leaks and anonymous sourcing. That means the policy remains provisional, and its final structure could still change.
What the reports say is in the draft
According to the reporting referenced in the source material, the latest version of the order is divided into two sections: one focused on cybersecurity and one focused on what it calls covered frontier models. The cybersecurity section is described as strengthening federal defenses against AI-related threats, which suggests an emphasis on protecting infrastructure rather than directly regulating private-sector model development.
The more consequential section for AI companies is the one dealing with frontier systems. Under the reported framework, developers would have a 90-day window to check in with the government and have a model vetted. What stands out is that this process is said to be voluntary.
That distinction could determine whether the order functions as a meaningful control point or more of a signaling mechanism. If companies can choose not to participate, then federal review becomes a matter of alignment and trust rather than compliance.
Why the apparent reversal is notable
Earlier descriptions of the administration’s AI posture pointed in different directions. A policy document released about two months earlier had given the impression of an unusually light-touch approach, with little beyond age restrictions for users. But subsequent leaks had suggested the executive order under development might move much further, potentially putting a government body such as the Center for AI Standards and Innovation in a position to vet new models before deployment.
That would have been a substantial turn toward intervention, especially from an administration publicly associated with skepticism toward regulation. The newly reported voluntary structure, if accurate, would therefore look like a partial reversal or compromise inside the drafting process.
That kind of internal tension would not be surprising. Frontier AI oversight raises competing pressures: national security concerns, industry lobbying, fears of slowing U.S. innovation, and questions about which agency should even be responsible for evaluation. A draft that appears to oscillate between tougher and softer approaches may reflect those unresolved battles rather than a settled philosophy.
The business logic problem
A voluntary review system creates an obvious incentive issue. Companies developing advanced models invest heavily to move quickly, differentiate themselves, and capture market share. Submitting a model for government vetting could introduce delay, disclosure concerns, and uncertain feedback. Unless participation brings a clear benefit, some developers may conclude that opting in is not worth it.
That is the practical criticism embedded in the reporting: if vetting is optional, the firms most worried about speed or secrecy may be the ones least likely to volunteer. Meanwhile, companies that already want to emphasize safety or policy cooperation may be more inclined to participate, even though they are not necessarily the ones policymakers are most anxious about.
Voluntary systems can still matter if they become de facto standards. For that to happen, however, government would need to create enough value in the process that nonparticipation looks risky to customers, partners, or the public. Without that, the framework may amount to an ask rather than an enforcement mechanism.
Cybersecurity may be the firmer piece
Ironically, the cybersecurity portion of the draft may end up being the more durable part of the order. Hardening federal systems against AI-enabled threats can often proceed through procurement rules, internal standards, and agency directives without directly confronting the politically fraught question of how to regulate private AI labs.
That could allow the administration to claim concrete action on AI risk even if the frontier-model provisions remain soft. From a policy perspective, though, that would leave the biggest governance question unresolved: how the federal government intends to monitor or shape the release of the most capable new models.
A policy signal, not yet a policy fact
For now, the most responsible reading is that the administration’s AI order remains in flux. The draft details circulating publicly suggest a softer landing than some earlier reports implied, with voluntary rather than mandatory sharing of frontier models for government review. But because those details come from leaks, they describe a process, not a final rule.
Even so, the direction of travel is important. If the final order preserves a voluntary framework, it will signal that the administration prefers influence, coordination, and infrastructure defense over direct compulsory oversight of model developers. That would be a consequential choice at a moment when the capabilities of advanced AI systems are rising faster than consensus on how to govern them.
The result could be an AI policy that sounds more watchful than restrictive: serious on cybersecurity, cautious on intervention, and still undecided about how much leverage Washington should actually exercise over the companies building frontier models.
This article is based on reporting by Gizmodo. Read the original article.
Originally published on gizmodo.com
