President Trump signed an executive order on Tuesday that demands frontier AI developers voluntarily submit their most powerful models to the government for security testing — but only gives regulators a maximum of 30 days to complete that review.
The order, titled "Promoting Advanced AI Innovation and Security," is a significantly slimmed-down version of an earlier draft that would have allowed up to 90 days for government review. Trump pulled that draft hours before a planned signing ceremony in late May, worried that a lengthy review window would throttle the pace of American AI development.
The final version makes clear that the Trump administration's AI policy remains firmly in "de-regulation, efficiency first" mode. Companies building cutting-edge models are asked — not required — to submit them for cybersecurity and national security testing. But the clock now runs fast, and the government must either clear the model or raise objections within 30 days.
The shift underscores a fundamental tension in AI policy: how to ensure safety without ceding competitive advantage. Trump's approach leans hard on speed, betting that voluntary, time-constrained oversight will keep U.S. models ahead of rival efforts.