On Tuesday, President Trump signed an executive order calling for the creation of a framework designed to give the federal government the ability to test AI models. The order is implemented by the Office of the National Director of Cyber, which is responsible for advising the president on cybersecurity issues, by creating a process that will allow the US to share information about software risks identified by AI systems such as Claude Mythos with operators of critical infrastructure, including banks, local services and hospitals, before those models are publicly available.
Trump was originally expected to announce the order on May 21, but according to Axios the White House postponed the signing ceremony following pressure from tech industry insiders. The president later told reporters that he “didn’t like certain aspects” of the original order. In accordance with PoliticsTrump participated in a small, high-level White House meeting where he and his advisers agreed on a new rollback plan. The new order, signed in a private ceremony, asks AI companies to share their powerful models for the government to review voluntarily 30 days before they are available to the public. An earlier draft had called for the government to be given up to 90 days to review the model, and some industry officials reportedly wanted that time shortened to 14 days before today’s announcement.
Ahead of the announcement, Engadget spoke with the Center for Democracy and Technology. “I think the idea of testing, especially for critical infrastructure providers, to be able to identify weaknesses and fix them before the capabilities are widely available, that makes sense,” said Samir Jain, the organization’s vice president of policy, told Engadget. Although he did not see the final order, Jain called the order “vague” at the time, noting that it did not give the public much visibility into the marking process.
“We don’t want a situation where any managers use countervailing power over when and how models are released, especially when they can use security as a pretense to block or disable a model for unrelated political or ideological reasons,” he said. “The opaque process allows for that possibility.”
That Trump has decided, after his previous reluctance, to regulate the AI industry in some way is a departure. In his AI Action Plan from last summer, the White House outlined a policy proposal that places few restrictions on OpenAI and others. So far when the president has sought to regulate the industry, he has done so purely for ideological reasons, issuing an executive order that restricted the federal government from purchasing “awakened” AI systems that “use responses favorable to DEI doctrines.” Trump has also sought to prevent states like Colorado and New York from overstepping their AI limits, even ordering the creation of an AI litigation group within the Justice Department to challenge state laws deemed “draconian” by the president.
“To the extent that there has been regulation, it has achieved ideological goals. It is fair to say that the Trump administration has been wrong about the risks and potential harms associated with AI,” Jain said. “In that sense, the executive order is a shift in how the administration sees AI as posing a real security risk and the government needs to take action to mitigate or address those risks.”
