Trump’s AI Order Is Strategic, Not Merely Deregulatory

By Jesse W. Lemon, Esq., CIPP/US, CIPP/E

 

On June 2, the Trump administration issued Executive Order No. 14409 on advanced artificial intelligence innovation and security, framing it as a national security and cybersecurity measure designed to strengthen federal systems, protect critical infrastructure, develop a classified benchmark for advanced AI cyber capability and create a voluntary early-access framework for covered frontier models with trusted government partners.

This article considers the strategic logic embedded within the order: Although the framework is styled as voluntary and innovation-friendly, it also creates a new mechanism for bringing the most capable AI systems into closer alignment with federal security priorities.

The AI order looks voluntary — that is the point.

The executive order establishes three central mechanisms relevant here.

First, it directs the development of a classified benchmarking regime to assess the cyber capabilities of frontier AI models.

Second, it creates a voluntary early-access framework through which developers may provide the federal government advance visibility into highly capable systems before public release.

Third, it integrates national security agencies, including the U.S. Department of the Treasury, into assessing systemic risk associated with broad deployment of these models. Although framed as voluntary and innovation-oriented, these features collectively embed federal visibility, influence and leverage into the lifecycle of frontier model development.

It does not announce a licensing regime. It does not require preclearance before frontier models reach the market. It does not create a federal board empowered to decide which AI systems may live or die.

Yet the order has teeth. They are not the teeth of a statute or a formal regulation. They are the teeth of access, trust, procurement, classified judgment and national security power.

The order’s significance lies not in what it compels, but in how it reshapes the incentives surrounding frontier AI development.

For the companies building frontier models, participation may be voluntary in the way much of Washington operates — not through overt command, but through access, process and the understood consequences of declining to engage.

That does not make the policy wrong. In fact, it may be the right structure for the U.S. to win the AI race, particularly with China.

But winning cannot mean releasing systems of strategic consequence into the world with no meaningful federal visibility.

The question is not whether America should move fast. It must. The question is whether it can do so with discipline and responsibility.

The first hidden tooth is the Treasury.

If this were primarily an AI standards initiative, the National Institute of Standards and Technology would naturally occupy center stage. If it were primarily an innovation initiative, the U.S. Department of Commerce would lead. If the administration’s principal concern were intelligence collection, one would expect the U.S. Intelligence Community to dominate the framework. Treasury’s role suggests something different.

The inclusion of the Treasury reflects a shift from evaluating discrete technical risk to assessing systemic exposure across interconnected sectors.

As an institution, the Treasury is responsible for assessing and responding to systemic risk. Its focus is not a single company, product or vulnerability. It is the possibility that a failure in one corner of the economy propagates through payment systems, financial institutions, insurers, cloud providers, utilities and other forms of critical infrastructure.

That perspective should sound familiar. Following the 2008 financial crisis, policymakers increasingly viewed risk as something that could emerge from the interaction of many actors, none of whom individually appeared dangerous.

The concern was not merely that a bank might fail. The concern was that interconnected systems could fail together. The Treasury’s inclusion in the order suggests a similar logic is beginning to emerge around frontier AI.

That does not mean the administration believes AI will trigger a financial crisis. It means the administration appears to be asking a different question than many participants in the AI debate.

Instead of asking whether a particular model is dangerous, it is asking what happens when highly capable models become embedded throughout the economy at the same time. That is the kind of systemic-risk question the Treasury is institutionally designed to address.

Once AI becomes a systemic-risk issue, the government possesses many more tools than direct regulation. It can work through financial institutions, insurers, procurement systems, critical-infrastructure operators and risk-management frameworks.

The Treasury has long operated through those forms of institutional leverage.

The second hidden tooth is the trusted circle.

The order’s voluntary early-access framework sounds unremarkable.

A frontier AI company may provide government access before public release. The government may identify trusted partners. The company may cooperate. Nothing on paper appears compulsory.

Participation remains formally optional. In practice, however, federal agencies, financial institutions, insurers and critical-infrastructure operators will treat participation as a signal. A model that enters the trusted process may become easier to purchase, insure, deploy, defend and integrate into government systems.

A model that remains outside the process may acquire a different reputation. No formal prohibition is required. The market can perform the discipline function.

This is how soft power becomes hard power.

This mechanism parallels federal approaches in cybersecurity and export control, where formally voluntary frameworks operate through market and regulatory signaling to produce de facto compliance.

The government does not need to announce that a model cannot be released. It needs only state that it lacks confidence regarding whether the model is appropriate for federal systems, critical infrastructure or trusted partners. That statement alone may alter commercial outcomes.

Over the last decade, administrations of both parties have increasingly shaped technology markets through procurement authority, cybersecurity directives, export controls, foreign investment review and critical-infrastructure governance.

Former President Barack Obama’s 2013 Executive Order No. 13636 directed a public-private framework for improving critical-infrastructure cybersecurity.

Former President Joe Biden’s Executive Order No. 14028 modernized federal cybersecurity through zero trust, software supply chain security, incident response and threat information sharing.

The Export Control Reform Act of 2018 authorized controls on emerging and foundational technologies critical to U.S. national security.

The Foreign Investment Risk Review Modernization Act of 2018 expanded Committee on Foreign Investment in the United States review of foreign investment in U.S. businesses, critical technologies, critical infrastructure, sensitive personal data and certain real estate transactions.

President Donald Trump’s 2019 Executive Order No. 13873 created a national security review framework for information and communications technology and services supplied by foreign adversaries.

Each of these developments reflect the same trend. The federal government increasingly influences strategic technologies not through direct licensing, but through access to markets, contracts, infrastructure, capital and trusted networks.

The third hidden tooth is the classified benchmark.

The classified benchmark may prove the most consequential feature of the order.

Public discussion has largely treated the benchmark as a sorting mechanism designed to determine which models fall within the early-access framework. But a classified measure of cyber-relevant AI capability is potentially far more significant than that.

Once capability assessments become integrated into national security decision-making, they can affect a broad range of government actions. Export-control determinations under the Export Control Reform Act, procurement decisions, intelligence priorities, cyber defense planning and technology-sharing arrangements with allies all depend on judgments about strategic capability.

A classified benchmark therefore risks becoming more than a measurement tool. It may become a governance tool. Once the government can classify model capability, it can govern around that classification. It can influence procurement. It can shape export controls. It can determine what allies may receive and what adversaries must be denied. It can inform intelligence collection priorities, cyber-defense planning, diplomatic negotiations and sanctions strategy.

The benchmark becomes a map of AI power. That is where the order becomes statecraft.

The U.S. is no longer asking only whether a model is safe. It is asking what the model can do, who should possess it, who should not and how American firms can remain technologically dominant without becoming detached from American security interests.

The order reflects a broader trend in national security governance. Increasingly, the federal government achieves strategic objectives through trusted relationships, procurement leverage, information sharing and market signaling rather than direct regulation.

Whether that trend is desirable remains open to debate. Whether it is real is increasingly

difficult to deny.

The framework works only if it is administered with discipline.

If Washington transforms the voluntary early-access and trusted-partner process into ideological review, compliance theater or a de facto permission structure, it will slow the very companies the country needs to lead. That would be a gift to China in the AI race.

But if the framework remains focused on frontier cyber capability, critical infrastructure, adversary access and national security risk, it can support innovation rather than smother it.

Innovation does not require government blindness. It requires speed, capital, talent and freedom to build. At the frontier, however, it also requires a state capable of protecting the conditions under which private innovation can survive.

China does not separate technology from national power. The U.S. cannot afford to pretend that frontier AI is merely another software market.

The Trump administration has not abandoned deregulation. Nor has it quietly adopted the Biden administration’s model under a different label.

The Trump administration’s order is not the former model with a red hat.

The order does not simply repudiate the Biden administration’s approach. The difference between the two is partly philosophical and partly temporal.

The Biden administration’s October 2023 Executive Order No. 14110 addressed AI as an emerging general-purpose technology that was beginning to spread across government, labor markets, consumer products, civil rights enforcement, education, healthcare and national security. Its breadth made sense for that moment.

The present order addresses models whose capabilities have concentrated at the frontier. The central question then was how to establish a federal governance baseline for a technology whose uses were multiplying faster than agencies could classify them.

The Trump administration’s framework responds to a later and more concentrated problem. The frontier has moved.

The policy question is no longer only how agencies should govern AI use across society. It is how the federal government should deal with models whose capabilities may matter for cyber offense, cyber defense, critical infrastructure, intelligence, military competition and the strategic contest with China.

That makes the two approaches different without making either one unserious. The Biden administration’s model governed through process: standards, reporting, agency management, civil rights review, consumer protection and federal oversight.

The Trump administration’s model governs through access: trusted early release, classified capability assessment, procurement leverage, critical-infrastructure channels, and national security market signaling.

The Biden administration treated AI as a societywide governance problem. The Trump administration treats frontier AI as strategic infrastructure.

The former administration’s framework asked how to make AI safe, secure and trustworthy across many domains. The current administration’s order asks how to keep America ahead while ensuring that the most capable systems do not remain invisible to the state.

That shift may reflect ideology. It also reflects time. AI in 2026 is not AI in 2023. Models are more capable, more widely deployed, more central to enterprise operations, and more plausible as tools of cyber and strategic competition.

A narrower, harder-edged national security framework may therefore be less a contradiction of the earlier approach than an adaptation to a more dangerous frontier.

The order’s real logic is strategic, not merely deregulatory.

The order preserves the public posture of voluntary cooperation while constructing a framework that sophisticated firms will find difficult to ignore.

That is the offer frontier AI developers cannot refuse: Come inside the trusted system, help the government understand what your model can do and remain part of the national project of winning the AI race. Stay outside, and accept the consequences that follow from being unknown, unvetted and strategically inconvenient.

The goal is to preserve American advantage in the AI race with China. The administration must do so without building a bureaucracy that crushes the firms that make victory possible.

The order reflects an effort to move quickly while preserving advantage, maintaining visibility into frontier capabilities, and integrating those capabilities into national security decision-making before external events narrow the available policy options.

The relevant question is not whether participation is required, but whether meaningful participation in the frontier AI ecosystem remains viable outside the framework the order constructs.