Trump Signs AI Executive Order: 30-Day Frontier Model Review

President Donald Trump on Tuesday signed an executive order asking artificial intelligence companies to give the federal government early access to their most powerful models for up to 30 days before release. The order, titled "Promoting Advanced Artificial Intelligence Innovation and Security," establishes a voluntary review process for what the administration calls "covered frontier models."

The signing ends weeks of speculation. A draft of the order was set to be signed in May but pulled at the last minute after Trump worried it would slow down American AI companies. The version signed Tuesday is notably shorter than the May draft, cutting the proposed government review window from 90 days down to 30.

Why this matters

The order is the clearest signal yet of how the Trump administration plans to balance AI safety with its broader competitiveness agenda. Throughout 2025 and early 2026, the White House has repeatedly framed regulation as a threat to US leadership in AI. Tuesday's EO is the first formal policy that asks AI labs to slow down at all — even on a voluntary basis.

The 30-day window also reshapes how companies like OpenAI, Anthropic, Google, and xAI will plan launches. A model the government wants to test for a full month can't be shipped on a surprise schedule. Product roadmaps, marketing cycles, and competitive timing all have to flex around it.

What was announced

The executive order creates four main mechanisms. First, AI developers can voluntarily provide the federal government with access to "covered frontier models" 30 days before public release. The administration will define the "covered frontier model" threshold using a classified benchmark, focused on the most powerful, high-energy models in production.

Second, the order establishes a voluntary AI cybersecurity "clearinghouse" coordinated by the Treasury Department and other federal agencies. Its job is to identify and remediate vulnerabilities in unreleased AI systems and to share information about emerging threats across industry and government.

Third, the order directs federal agencies to develop benchmarks for evaluating AI models' cyber capabilities — both offensive and defensive — and to use those benchmarks to shore up the government's own security defenses.

Fourth, the EO expands hiring at the US Tech Force and gives the federal government a role in selecting the "trusted partners" that get early access to frontier models alongside government reviewers.

Industry impact

For frontier labs, the most immediate question is whether participation in the voluntary program will become de facto mandatory. Refusing to share a model with the government — when competitors do — could become a procurement liability for federal contracts, even though the EO has no explicit teeth.

OpenAI and Anthropic have both told reporters they support the framework in principle. Both companies already do extensive pre-release safety testing and have existing relationships with US AI safety institutes. The 30-day window aligns roughly with what their internal red-teaming cycles already take. xAI and Meta have been quieter publicly, but both face higher friction: xAI's Grok models ship on aggressive timelines, and Meta releases models openly, where a "pre-release" review concept doesn't map cleanly.

For enterprise customers, the cybersecurity clearinghouse is the more practical piece. A central federal coordinator for AI vulnerability disclosure gives buyers a clearer escalation path when something goes wrong with a deployed model. That matters more in regulated industries like finance, healthcare, and defense than the headline 30-day review.

Expert perspectives

The Scientific American analysis called the order "a drastic shift" in the administration's stance, noting that earlier Trump executive orders had been almost entirely about removing AI guardrails rather than adding them. The Council on Foreign Relations was more measured, framing the EO as "the first serious oversight framework" the administration has produced for advanced AI.

Cybersecurity reporters noted that the clearinghouse is the underrated piece. Putting Treasury at the center of frontier AI security coordination signals that the administration sees AI risk as a systemic financial and infrastructure problem, not just a research question. That framing aligns with how regulators treat critical infrastructure software.

Civil liberties groups raised concerns about the classified benchmark used to define a "covered frontier model." Without public visibility into the threshold, AI labs and outside observers can't independently evaluate which systems trigger the review or what the government is testing for.

What's next

Treasury and the other coordinating agencies have 90 days to stand up the cybersecurity clearinghouse and publish initial vulnerability disclosure procedures. The classified frontier model benchmark is due within 120 days. AI labs will likely begin informal conversations with government reviewers well before the formal program is live, especially companies with major model launches planned for late 2026.

Watch the procurement angle. The Department of Defense and the intelligence community are the most likely first movers in tying participation in the voluntary review to federal contract eligibility. If that happens, the program becomes mandatory in everything but name for any lab serving the US government.

Also watch Congress. The EO does not preempt legislation, and several pending bills on AI safety and licensing remain in committee. A formal executive framework may either accelerate those bills (by setting a baseline Congress wants to expand) or stall them (by letting members claim the issue is being handled).

Bottom line

The Trump administration just put its first real AI safety policy on paper, and it's a voluntary 30-day review of the most powerful models before they ship. It's lighter touch than the May draft, but it changes how launches get planned and gives the federal government a structured seat at the table. Watch whether participation becomes a procurement requirement — that's the moment "voluntary" stops meaning voluntary.