Trump Signs AI Executive Order: Frontier Models Get 30-Day Government Preview Before Release

President Trump signed an executive order on June 2 titled “Promoting Advanced Artificial Intelligence Innovation and Security” that asks AI companies to voluntarily submit their most powerful models to the federal government for up to 30 days of review before any public release. The order also creates an AI cybersecurity clearinghouse and directs agencies to develop benchmarks for assessing frontier models’ offensive cyber capabilities.

The word “voluntarily” is doing a lot of work here. The original draft of this order was pulled weeks before the signing specifically because it gave the government 90 days to review frontier models — a timeline the White House judged would chill innovation. The final version cuts that to 30 days and removes any coercive mechanism: companies can decline without penalty. The order explicitly bars the government from creating a mandatory licensing or preclearance requirement.

The AI cybersecurity clearinghouse is the more concrete piece of the policy. Federal agencies will share information on vulnerabilities discovered in AI systems, and the benchmarks program will help the government independently assess what frontier models can actually do on offense. That capability gap has been a persistent source of policy anxiety since GPT-4-class models started assisting security researchers in finding zero-days without being specifically trained to do so.

For AI labs, accepting the arrangement is straightforward. Thirty days of optional government access costs almost nothing: the government lacks the workforce to meaningfully audit a frontier model anyway, and participation can be framed as patriotic cooperation. What the order establishes is the infrastructure for a closer government-industry relationship that future administrations or Congress could formalize into something binding.

The contrast with the EU AI Act is sharp. The European framework mandates conformity assessments for high-risk AI systems and requires general-purpose AI model providers to publish technical documentation and comply with copyright law. The U.S. approach under this order is softer: coordination over compulsion, industry self-governance with light government oversight.

The White House indicated separately that the AI cybersecurity clearinghouse will operate through CISA — the same agency that manages the Known Exploited Vulnerabilities catalog and oversees federal civilian cybersecurity. Extending CISA’s mandate into AI model capabilities is a logical structural move. Whether that mandate will have real teeth depends on what happens after voluntary cooperation produces its first data.

Sources

• CNBC
• NPR