The Trump administration signed an executive order on Monday titled "Promoting Advanced Artificial Intelligence Innovation and Security" that establishes a voluntary framework for the federal government to review frontier AI models before their public release.
The word "voluntary" is doing all the work.
What happened in May
This order was supposed to be signed weeks ago. The White House pulled it after industry pushback centred on a 90-day government review window that AI companies argued would slow them down in the race against Chinese competitors. The signed version cuts that window to 30 days.
The earlier draft spooked labs because a three-month hold on model releases would hand a structural advantage to rivals operating outside US jurisdiction. Thirty days is still friction, but it is friction the industry can absorb, and Wedbush analyst Dan Ives described the final version as "incrementally positive" for the sector.
What the order does
AI companies are asked, not required, to provide frontier models to the federal government up to 30 days before release. A classified benchmarking process will determine whether a model qualifies as a "covered frontier model," replacing the Biden administration's static compute threshold of 10^26 floating-point operations.
The order creates a voluntary AI cybersecurity clearinghouse to coordinate vulnerability detection and patching across government and critical infrastructure. It directs agencies to harden federal and critical-infrastructure systems on 30 to 60-day timelines. It expands federal hiring for cyber and AI roles and prioritises enforcement against AI-enabled cybercrime.
The order also disclaims any mandatory licensing, permitting or pre-clearance for model development or release. That distinction is deliberate and aimed at the industry's core concern: that review would become approval, and approval would become a gate.
The government as customer
The more consequential shift is structural. By requiring agencies to evaluate frontier models, the order creates a procurement funnel. The government becomes a buyer of the models it reviews, opening a public-sector revenue stream that is less cyclical than enterprise sales and carries the implicit endorsement of federal adoption.
For companies like Anthropic, OpenAI and Google, government contracts provide stable, high-margin revenue and a credibility signal that accelerates enterprise adoption in regulated industries like healthcare, finance and defence.
Wedbush noted that the cybersecurity clearinghouse will also benefit AI-native security vendors as enterprise security budgets accelerate in response to increasingly sophisticated AI-enabled threats.
The gap between voluntary and mandatory
The order is framed as a partnership. The government asks to see the models. Companies can say no. There are no penalties for declining.
In practice, the line between voluntary and expected is thin. A frontier AI company that refuses to participate in a government review programme, while its competitors comply, creates a reputational and commercial disadvantage for itself. The framework is voluntary in the same way that airport security is optional: you can decline, but you will not be boarding the plane.
The administration pulled this order back once. It returned softer, faster and more industry-friendly. The message to AI companies is clear: we are not here to stop you. We are here to buy what you build.
Whether that remains the posture when the first serious AI incident hits critical infrastructure is the question the order does not answer.
