President Donald Trump signed an executive order Tuesday establishing a voluntary framework for artificial intelligence companies to share their most advanced models with the federal government before public release, a move aimed at bolstering national cybersecurity without imposing the kind of mandatory licensing regime that critics feared from earlier drafts.
The order gives the government access to what it calls "covered frontier models" up to 30 days before broader release, pending federal review. That window is a sharp reduction from the previous version of the order, which would have required up to 90 days of review access.
Trump signed the order privately, nearly two weeks after postponing a planned signing ceremony with technology industry executives. At the time, the president said he "didn't like certain aspects of it." The final version reflects those concerns, and draws a clear line against government gatekeeping over AI development.
The executive order directs five agencies and offices, the Treasury Department, the National Security Agency, the Cybersecurity and Infrastructure Security Agency, the National Institute of Standards and Technology, and White House officials, to build a classified benchmarking process within 60 days. That process will evaluate the advanced cyber capabilities of AI models.
Those same officials must also define the threshold at which a model qualifies as a "covered frontier model", the trigger point for voluntary pre-release government access.
War Secretary Pete Hegseth was among the administration officials directed to take action under the order, though the specific tasks assigned to him were not detailed.
The voluntary nature of the system is its defining feature. The order states plainly:
"Nothing in this section shall be construed to authorize the creation of a mandatory governmental licensing, preclearance, or permitting requirement for the development, publication, release, or distribution of new AI models, including frontier models."
That language matters. It forecloses, at least for now, the possibility that this executive order could morph into a federal permission slip for AI deployment.
Two figures reportedly played influential roles in ensuring the order prohibited mandatory government licensing: David Sacks, the former White House AI czar who now co-chairs the President's Council of Advisors on Science and Technology, and Ryan Baasch, the National Economic Council's deputy director.
Their fingerprints on the final text suggest an internal push to keep the administration's AI posture pro-innovation. The earlier draft's 90-day review window and broader scope had raised alarm bells in parts of the tech industry. The final order's 30-day window and explicit ban on mandatory licensing represent a significant concession to those concerns.
The ground had already been softened. Last month, several Big Tech companies, including Google and Microsoft, agreed to provide the federal government access to their AI models for national security purposes. The executive order formalizes and extends that cooperation, but does not compel it.
The distinction between voluntary and mandatory is not academic. It is the difference between a government that partners with industry to identify genuine security threats and one that inserts itself as a gatekeeper over which technologies reach the market.
The Biden administration's approach to AI regulation leaned heavily on the latter instinct, toward mandates, disclosure requirements, and bureaucratic review layers that threatened to slow American innovation while competitors in Beijing moved without such constraints. Trump revoked Biden's AI executive order early in his second term, and this new order charts a fundamentally different course.
A classified benchmarking process does raise fair questions. Who decides what counts as a "covered frontier model"? What happens if the 60-day deadline for defining that threshold slips? And what leverage, if any, does the government have over companies that decline to participate in a voluntary system?
Those questions remain open. The order sets timelines and assigns responsibilities, but the real test comes in execution.
Cutting the pre-release review period from 90 days to 30 was more than a cosmetic change. In the AI industry, three months is an eternity. Models evolve fast. Competitors release fast. A 90-day government hold on a frontier model could hand a decisive advantage to foreign rivals operating under no such constraint.
Thirty days is still a meaningful period, long enough for federal cybersecurity experts to probe a model's capabilities, short enough that companies are not penalized for cooperating.
The administration appears to have landed on a framework that takes national security seriously without treating American AI firms like suspects. That balance is worth preserving.
The 60-day clock starts now. The Treasury Department, NSA, CISA, NIST, and White House officials must produce a classified benchmarking process and define the "covered frontier model" standard within that window. The criteria they set will determine whether this order functions as a lightweight security check or becomes something heavier.
The involvement of agencies ranging from Treasury to the NSA signals the administration views AI's security implications as spanning both economic and intelligence domains. That is a realistic assessment. Advanced AI models can be weaponized for cyberattacks, financial manipulation, and intelligence exploitation. Getting a look at the most powerful models before they go live is a reasonable precaution, provided the government earns industry trust rather than demanding compliance.
Google and Microsoft have already shown willingness to cooperate. Whether smaller AI firms and newer entrants follow suit will depend on how the benchmarking process is structured and whether participation carries any hidden costs.
The administration deserves credit for resisting the regulatory reflex. Voluntary cooperation, a short review window, and an explicit prohibition on mandatory licensing, that is a framework built for a country that wants to lead in AI, not just regulate it.
Washington's instinct is always to add a permission slip. This order, at least on paper, says the opposite.
