Trump signs an executive order to vet top AI models for national security risks

WASHINGTON — President Trump signed an executive order Tuesday directing the federal government to establish a voluntary early review process for the country’s most advanced artificial intelligence models, following a months-long internal battle over how aggressively Washington should move to regulate the fast-growing technology.

Under the order, companies are asked to allow government agencies, including the National Security Agency and representatives of the Defense Department, to evaluate cutting-edge models up to 30 days before they are released to the public. The order stops short of mandating participation and explicitly bars the creation of any new licensing or permitting for AI models.

“The main question is whether this is the start of a continued government clamp down and response to continued AI capabilities, or whether this is a one-off, limited, and truly voluntary act,” said James Sanders, research associate at the Center for a New American Security, a Washington, D.C., think tank.

“It’s unclear how voluntary this will stay and how voluntary it will be in practice as the AI labs try to maintain good relationships with the U.S. government,” he said.

The order represents a reversal for Trump, less than two weeks after he scuttled a version of the policy that gave the government a 90-day review period — and, more broadly, for an administration that came to power promising to strip away AI guardrails, a posture that slowly created fractures within the GOP.

In the executive order, Trump appeared to frame a need to foster AI technologies while taking into account national security. “As these capabilities evolve, my Administration will continue to work closely with industry to ensure that the best and most secure technology is deployed rapidly to confront any and all threats to our country,” he said in the order.

The step set off immediate debate about whether Trump’s plan would be an effective approach. It formalizes an existing practice in which top AI companies share models with external evaluators and government players before deploying them publicly, but raises questions about how voluntary it will be and how the government will choose which labs to target.

David Sacks, who previously served as Trump’s AI advisor, called the 30-day window a “game changer,” arguing that the shorter timeline would allow companies to engage with the government without slowing down new model releases.

“In the AI race, every day counts,” Sacks wrote in a post on X.

Dean W. Ball, Trump’s former AI advisor, characterized the order as a victory for the AI “safety contingent” and a loss for Sacks and others who promote a more accelerated approach. He called the order a mistake, saying it could be a first step toward a federal licensing requirement for AI models.

“All for a benefit that is barely articulable; what, exactly, is the intelligence community going to do in 30 days to make the models safer?” Ball wrote on X.

The signing of the executive order occurred amid growing tensions among Republicans over AI, job loss and data center construction, including fear among a significant portion of Trump’s supporters that artificial intelligence could eliminate jobs or become a security threat. Polling in May had shown strong support among Republicans for a framework like the one outlined in Trump’s executive order.

The growing split among Republicans over AI was clearly visible in Florida on Monday, where James Uthmeier, the state’s Republican attorney general, sued OpenAI over the alleged risks of ChatGPT, citing the use of the bot by a gunman in a shooting at Florida State University last year.

Meanwhile, Rep. Byron Donalds — the Trump-endorsed candidate to succeed Gov. Ron DeSantis — said Monday that he did not agree with Trump on AI policy, indicating he supported state-led regulation, a shift for a candidate who had been backed by the AI industry earlier in the year.

A poll released by Americans for Responsible Innovation, a nonprofit advocating for a federal framework for AI policy, found that the majority of Republican voters polled supported the type of plan laid out in Trump’s executive order. Seventy-one percent also said independent security testing should be required by law for advanced AI systems.

When Trump took office, his administration pivoted away from Biden-era policies requiring AI companies to test their AI models and share safety results with the government before public release, reversing the U.S. posture on regulation.

That changed after Anthropic — acting on its own initiative — brought its Claude Mythos Preview model to senior White House officials, a move that exposed vulnerabilities in its software and raised concerns about the potential need for safety-testing of AI models before broad public release.

The White House attempted to downplay the executive order as a regulatory move, emphasizing in a post Tuesday that the federal government would not conduct sweeping oversight and the process outlined in the executive order would be voluntary.

“We are NOT conducting oversight of all new models, as that level of government overreach would have chilling effects on free speech and innovation,” the White House Office of Science and Technology Policy posted on X.

Trump’s signing of the order prompted calls from those who support stricter AI regulation for Congress to take steps beyond Trump’s plan. Thus far, Congress has not passed any major legislation to regulate artificial intelligence.

“Congress should take the structure this order creates, make participation mandatory, and extend it beyond cyber threats to the full range of risks the most capable models present,” Riki Parikh, policy director of the Alliance for Secure AI, a nonprofit that promotes safeguards for AI, said on X, saying the order’s voluntary framework “isn’t enough.”

Progressives, including Gov. Gavin Newsom and Vermont Sen. Bernie Sanders, said the executive order was too weak and slammed Trump for flip-flopping on regulation.

Some experts suggested the distinction between voluntary and mandatory sharing of their cutting-edge technology may be crucial.

“No company is formally required to participate, but if a developer wants to sell frontier AI systems to the federal government, participation may soon become the price of entry,” Jessica Tillipman, a professor who studies contracting law at George Washington University, wrote in a post on X.

The administration’s approach was welcomed by industry leaders, including Microsoft President Brad Smith, who said the order was “an important step toward advancing innovation while protecting the security of the American public.”

Anthropic endorsed the order and called it “an important step in strengthening America’s leadership in AI.” The company said it was looking forward to supporting the implementation of the program.

Ceballos and McDaniel reported from Washington, Christopher from Los Angeles. Times staff writer Michael Wilner contributed to this report.