President Donald Trump’s recent executive order on artificial intelligence is being treated as a “shift” in the administration’s approach to tech regulation. It is welcomed as a cautious step toward responsible risk regulation, hailed as a qualified victory for tech critics, and framed as the outcome of “months of debate” within the administration over important matters of principle.
But this story misses the important ways that the executive order marks a continuation of the Trump administration’s preferred mode of governing. Trumpian regulation is premised on broad executive discretion, bypassing ordinary rulemaking procedures and, as far as possible, judicial accountability. Where the coercive power of the federal government is insufficient, Trumpian regulation works by building strategic alliances with self-interested private sector elites. And it is premised on control over the circulation of goods, funds, and information.
These all-too-common regulatory techniques have roots far deeper than the current presidency, but they are particular hallmarks of the second Trump administration’s approach to executive power. And they are all on display in the new AI order. Viewed in this way, the order is not a “pivot” for the Trump administration, but just the latest page in the same playbook.
Coordinated, Voluntary Self-Regulation
The president’s June 2026 order on AI purports to avoid burdensome “top-down” regulation, and instead to work “hand-in-hand with American industry to strike the right balance between innovation and national security.”
The order accomplishes this goal by adopting a paradigm of voluntary and coordinated industry self-regulation. The word “voluntary” appears twice in the order, both times in its key operative provisions. First, the order establishes an “AI security clearinghouse,” which is meant to foster “voluntary collaboration with the AI industry and operators of critical infrastructure” on software vulnerabilities and their remediation. Second, the order contemplates a “voluntary framework” for the review and sharing of so-called “frontier models”—a bit of tech jargon with its own sketchy origin story.
The order makes clear that it does not create any kind of mandatory licensing or preclearance framework. But the administration could not have lawfully done such a thing without either invoking some existing statutory authority (which the order does not do) or getting Congress to pass a new law. Any attempt to implement such a permitting process, without plausible statutory backing, would be immediately frustrated by the courts. Even a Trump-aligned Supreme Court majority—struggling with its own legitimacy crisis—would not rush to bless a new regulatory process governing a massive and growing industry, if that process lacked even an arguable statutory basis.
But the fact that regulation is “voluntary” does not make it meaningless. Industry frequently commits to voluntary regulations, usually for a range of self-interested reasons. The key, then, is understanding why industry might be drawn into signing up. And, in the second Trump administration, experience tells us to look not at how industry will be restricted, but how both industry and the White House, together, will be empowered by the new order.
The New Security Framework
Perhaps the most important provisions of the order concern the regulatory process for new technology under section 3, framed by the order as “Secure Frontier Model Deployment.” Key elements in this process are undefined or inscrutable, and, as always, it is dangerous to act as if a Trump order makes any kind of clear sense. But an ambitious application of the order would empower the White House, in cooperation with federal agencies and a small set of AI developers, to restructure and control the market for emerging technologies.
The actual contours of the review procedure contemplated by the order are as follows. At an initial stage, the National Security Agency (NSA), in consultation with other agencies, will conduct an “assessment” of “the advanced cyber capabilities” of an AI model under development. On the basis of this assessment, the NSA will “determine” whether the model qualifies as a “covered frontier model.” This assessment and designation will be conducted according to a “benchmarking process.” The details of the process will be “classified,” though the assessments themselves may be shared “with AI developers and researchers as appropriate.” AI developers are also expected to “engage the Federal Government to determine whether model(s) under development” meet the criteria for designation.
The consequences of designation appear to follow a two-step sequence. First, the “Federal Government” will be given access to the model for a 30-day period. The purpose of this 30-day access period is not expressly stated in the order, but press coverage suggests it is “to give officials time to identify and address potential vulnerabilities before they are released publicly.” In the public reporting, the length of this period was said to be a major sticking point, and cutting the time from an originally planned 90 down to 30 days was said to be critical for mollifying industry skepticism.
Second, something happens at the end of that 30-day period, but what happens is not entirely clear. The order expects that AI developers will give designated models to the government 30 days “before they plan to release such models to other trusted partners.” The term “trusted partner” is never defined; the order simply says that developers will “collaborate” with the government to “select trusted partners that will have early access to covered frontier models.” The order also never says how long this “early” access will last, nor does it provide parameters for when and under what terms anyone who is not a “trusted partner” can access the model. Nor is it clear what happens to the government’s access at the end of this 30-day period: presumably, government access could cease, or the developer might agree with the government to extend its access beyond that period. And it is at least possible that the government and a developer might never decide to release a model broadly to the public, keeping the model instead under the control of the government and a few “trusted partners.”
The implications here are stark. The NSA, acting according to a secret procedure, can designate any emerging AI technology as “covered frontier models.” Once designated, the Executive Branch can collaborate with AI developers to keep that model controlled within a circle of “trusted partners,” who may enjoy privileged access for an indefinite period. Suggestions in the press that the order simply provides for a 30-day review period before models are released “to the public” are either mistaken, or they have an idiosyncratic understanding of what the “public” means.
If industry signs up to this procedure, then it provides a mechanism for potentially indefinite control of the market. But that means all this depends on the willing cooperation of industry. So, what’s in it for them?
Coordinated Regulation in the Chokepoint State
In forthcoming work, I argue that the second Trump administration has consolidated a mode of governing I call the “Chokepoint State.” Political scientists Henry Farrell and Abraham Newman, in their landmark study of networked power, describe the “chokepoint effect” as the ability of privileged states to control third parties’ access to crucial nodes in economic or information networks. In the Chokepoint State, I argue, administrative capacity is increasingly arranged around controlling these nodes in order to reward allies, punish enemies, and set rules.
The techniques of the Chokepoint State are not new, but their uptake and proliferation have accelerated under Trump. The AI executive order itself, for example, resembles both Trump and Biden-era chokepoint policies in some ways. By attempting to ring-fence the market for emerging technology, the new executive order reflects the “small yard, high fence” strategy of using export controls to structure the global market for semiconductors. And by offering “early access” to “trusted partners,” the order mimics the Compact for Higher Education’s offer of privileged “access” to federal funding in exchange for accepting new quasi-legal obligations on a wide range of issues.
The difference, of course, lies in the AI order’s explicitly voluntary nature. U.S. export controls rely on a web of regulations, statutes, and (for a large part of their history) emergency powers that enable the Executive Branch to control foreign commerce under the threat of civil and criminal penalties. The Compact for Higher Education, though formally voluntary, relied on explicit and implicit threats that leveraged the Executive’s power over spending, grantmaking, and visa approvals. Here, the new executive order invokes no statute and disclaims any pretension to impose a “mandatory” regime.
The order thus shows how voluntary and coordinated self-regulation might work in the Chokepoint State. The leading tech firms, having leveraged billions of dollars on AI, would naturally seek U.S. government protection from forces that would erode their dominant position. Under these conditions, it makes sense for these firms to seek a partnership with the White House. Under the framework of this order, or a similar regulation, the leading firms can establish joint public/private control over advanced AI technology, rebranding industry leaders as “trusted partners,” and entrenching their market and political power.
This perhaps accounts for the muted reaction from AI moguls. OpenAI founder Sam Altman posted a brief positive endorsement on X. And, the same day as the executive order, his company released a nine-page policy paper on the governance of “frontier AI.” The paper diverges from the executive order in many of its details, indicating that Altman & co. view the final shape of the administration’s regulatory process as open to negotiation.
But the continuities are more striking than the differences. OpenAI’s proposals also imply the existence of a network of U.S.-approved trusted partners. One executive told Politico that OpenAI “plans to push policymakers in Washington to create mandatory evaluation processes for AI companies that develop advanced systems,” adding, “[w]e don’t think any specific lab should be making that decision unilaterally.” As these statements suggest, cooperative, voluntary regulation allows the major players to borrow the legitimacy of the state, positioning themselves as the adults in the room while guarding the door to keep out potential competitors.
In this political-economic context, a process built on ad hoc executive discretion may suit insiders just fine. Regulation by executive order, discretion, and “dealmaking”—often just a euphemism for corruption—is the preferred mode of this administration. And, so long as they are confident of their access and influence, large private-sector players stand to gain more from shaping executive discretion than they might get from comprehensive federal regulation. Converting this process into a mandatory regime would only further lock in these dynamics, leading to what one firm has warned could be “a tiered system” in the market for emerging technology.
Meanwhile, smaller players will struggle to catch up. One of the more insightful analyses produced in the first twenty-four hours after the order came from the law firm A&O Shearman. The authors note the complete absence of criteria for identifying “trusted partners.” And they urge companies to engage “proactively” with relevant agencies to position themselves favorably “for trusted-partner designation.” Although it may be coincidence, it is not shocking that the lawyers at A&O Shearman—one of the firms that capitulated to White House bullying in 2025—know a patronage relationship when they see one.
Beyond rewarding allies and punishing adversaries, the group that controls a chokepoint can also set lasting rules of the road. This is a self-evident purpose of the new executive order. In addition to the secret rules for making “frontier model” designations, the order also expects that the government-review and trusted-partner procedures will be part of a “voluntary framework” that is designed by Treasury and other agencies “with AI developers.” No other group is mentioned.
The result is that the rules for deciding who gets “early” (likely meaning “privileged and indefinite”) access to state-of-the-art technology will be devised by, and for the benefit of, a small group of self-interested firms and their partners in the administration. And these rules may develop largely outside the procedures required of legislation or administrative rulemaking.
Beyond State versus Market
These important dynamics are missed and obscured by framing the new order as a step toward “regulation” of AI. And this framing is pervasive. Politico trumpeted the new order with the headline “Trump’s AI order is a blow against laissez-faire,” and quoting such “AI safety hardliners” as Steve Bannon. This framing poses a stark alternative between government “intervention” and a free market in which innovation occurs spontaneously.
But we have never been laissez-faire. The relevant question is not whether regulation takes place, but how a regulatory scheme shapes the distribution of public and private power. By placing significant discretion in the hands of an increasingly unaccountable executive, the Chokepoint State obviously enhances the power of the White House. At the same time, executive leverage over chokepoints can enhance the power of private actors who enjoy privileged access to the Oval Office.
The latest AI executive order may come to nothing. If anything can be said with certainty about this administration, it’s that a great many Trump initiatives start with a big media splash before dying a quiet and unceremonious death. But by providing a platform for close cooperation between the White House and industry insiders to control access to emerging technology, the order shows off this administration’s predilection for the techniques of the Chokepoint State.
