The federal government has spent months debating how much oversight — if any — it should have over the AI industry’s most powerful systems. President Trump signed an executive order Monday, June 2, that offers one answer: ask nicely.
The order
The order directs AI companies to voluntarily submit their most capable models for government testing up to 30 days before those models go public. It doesn’t mandate participation. Companies that choose to comply would hand over access ahead of release; those that don’t face no stated penalty under the current framework.
Beyond the pre-release review window, the order instructs federal agencies to develop benchmarks for assessing AI models’ cyber capabilities — a concrete technical task, though the timeline for completing those benchmarks wasn’t specified in the order. Agencies are also directed to stand up what the administration is calling an “AI cybersecurity clearinghouse,” a channel for reviewing and sharing information on software vulnerabilities tied to AI systems. The government’s own security defenses are targeted for upgrades as well.
The voluntary nature of the submission process is the sharpest edge of the debate here. Critics of a lighter-touch approach have argued that self-policing by companies racing to ship the next model isn’t enough — that meaningful safety review needs teeth. The order, as signed, doesn’t provide them. Whether major AI developers will actually hand their models over for a 30-day federal look before launch, or treat the ask as optional in practice, remains an open question.
The administration has not said which agency would lead the testing process or how the government plans to staff and fund reviews of models that can take teams of specialists weeks to evaluate. NPR hasn’t reported a response from any major AI company to the order as of Monday, June 2.
Reported by NPR. Read the original report.
