Trump AI order creates 30‑day voluntary review for frontier models

This order is Washington’s latest attempt to get ahead of frontier AI without slamming on the brakes. By asking—rather than forcing—labs to hand over powerful models up to 30 days before launch, the administration is signaling that cyber and national‑security risks from AI have crossed a threshold that justifies some level of pre‑deployment scrutiny. At the same time, the explicit ban on mandatory licensing and the relatively short review window reflect intense lobbying from industry, which feared a de facto slow‑roll of new models.

For the race to AGI, the symbolism matters almost as much as the substance. The U.S. federal apparatus is now building a permanent “AI cybersecurity clearinghouse” and a benchmarking regime for what counts as a covered frontier model. That pushes governments deeper into technical model evaluation and creates a template other countries can copy or harden. In practice, voluntary 30‑day reviews are unlikely to materially slow leading labs, but they do normalize the idea that ultra‑capable models must be inspected before wide release. Over time, as incidents accumulate, that voluntary baseline could quietly evolve into something much closer to binding oversight.