Trump's AI Order: A Win for Skeptics, But Is It Enough?

Even more striking is that the order came from a Trump administration determined to beat China in the race for AI supremacy — and to eliminate red tape that could get in the way.

“They have exploded the Overton window,” said Brad Carson, a former Democratic congressman and president of Americans for Responsible Innovation, a nonprofit backed by a faction of tech billionaires that pushes for AI rules. Like others in that camp, Carson argued that the White House’s new, voluntary approach to vetting — a 30-day process through which top AI firms agree to share advanced models with the government in order to identify and address cybersecurity risks — gives momentum to the so-called AI safety movement.

Steve Bannon, the far-right former Trump adviser and advocate for AI regulations, called Tuesday’s order a win for conservative skeptics of Silicon Valley.

“For the first time it’s on a piece of paper, a structure and a process,” Bannon told POLITICO. “That process is still pretty ill-defined, it doesn’t meet our requirements. But as I tell people, we’re going to eat the elephant one bite at a time.”

Even a voluntary AI vetting framework, he said, “implies it’s going to be higher cost for [the AI companies] if any of their models are wrong or have huge problems, and they let it out without bringing it in for voluntary review.”

“It’s going to raise the stakes for them,” Bannon said. He was part of a coalition of more than 60 conservative activists and Trump allies who signed onto a letter last month urging the president to require mandatory vetting.

Saif Khan, a former adviser on emerging technology in the Biden administration and a fellow at the Institute for Progress think tank, also believes Trump’s new order could ease the path for tougher AI rules down the line.

“Directionally, this is the kind of thing that I think many in the AI safety community would want to see,” Khan said. Despite language at the top of Tuesday’s order trashing former President Joe Biden’s “burdensome” approach to AI rules, he said Trump’s new AI safety plan is functionally identical to his predecessor’s.

“This executive order is implementing a voluntary regime to do pre-deployment evaluations of models for security risks,” said Khan. “And that is the thing that the Biden administration was doing.”

Caleb Knapp, senior policy manager at the Alliance for Secure AI, called the order “a good start toward building the institutional capacity needed for substantive oversight of advanced AI models.” But he said a voluntary framework “is not enough,” and contended that the White House has now provided Congress with “both an opportunity and responsibility to take this directive one step further and require AI developers to share their models for review before release.”

Many people in and around the AI industry also said the new White House order could signal the first step toward tougher rules on the technology. But unlike AI safety advocates, they view the possibility with trepidation.

“Rarely does the government do things that are only voluntary,” said Caleb Max, president and CEO of the National Artificial Intelligence Association, a trade group representing the AI industry. “These things typically only get tighter, not looser.”

Patrick Hedger, director of policy at the tech trade association NetChoice, also said that Tuesday’s order cracks open the door for tougher AI rules.

“While we anticipate President Trump’s new Executive Order to be a light-touch and collaborative approach as long as President Trump is in office, NetChoice is concerned that future administrations won’t honor President Trump’s vision,” Hedger said in a statement. He added that his organization is “concerned the collaborative framework President Trump has put forward will not remain voluntary.”

In a statement to POLITICO, White House spokesperson Liz Huston called Trump “the most pro-innovation President in American history” and said the order “reflects his common-sense approach of collaborating with industry to balance innovation and security, cementing America’s continued global dominance in AI and cybersecurity.”

The executive order comes about two weeks after Trump abruptly pulled plans to sign a nearly identical document. That about-face followed intervention from David Sacks, a venture capitalist and the president’s former AI czar who warned Trump that the order could chill AI innovation.

Trump said at the time he was worried the original directive would harm competition with China, telling POLITICO in an interview that he had many concerns about the draft order. But Tuesday’s order differs from the prior version in just one key respect: It reduced the timeline to vet advanced AI systems from 90 days to 30 days, a change intended to ensure that new model releases aren’t held up in months of reviews.

In a social media post, Sacks called the new 30-day timeline “a game changer.” But Dean Ball, a former White House adviser on AI and primary author of the Trump administration’s AI Action Plan, called it a “very small” concession in his own social media post — where he also worried the order would set the stage for government licensing of AI models.

Several people also questioned whether the new vetting framework is truly voluntary, given that reviews of advanced AI models will be conducted by the National Security Agency and other defense agencies. Such reviews were previously handled by the Commerce Department’s National Institute of Standards and Technology.

“It makes so much more sense to have a civilian agency be the front door for something remotely resembling domestic regulation,” said Khan, the former Biden official. “While I think it is valuable for the intelligence community to be involved in research on cybersecurity evaluations, I don’t think you want the intelligence community to regulate domestic industry, and that’s what this starts to verge on.”

Joseph Hoefer, principal and chief AI officer at the bipartisan lobbying firm Monument Advocacy, said he also anticipated some anxiety from the industry over another aspect of the order: a classified “benchmarking” process in which agencies will assess the national security implications of new AI models and determine which ones the policy covers.

“Developers are being asked to engage with a threshold they can’t fully see,” he said. “Most will accept that for national security reasons, but the labs that pride themselves on predictability are going to want as much clarity as the government can give them about where the line sits.”

Still, several tech lobbyists who spoke to POLITICO on Tuesday suggested they are broadly happy with the new order — provided that it doesn’t lead to significantly stricter rules down the line.

“I don’t see this as a zero-sum game,” said Paul Lekas, head of global public policy and government affairs at the Software and Information Industry Association. “We reject the ‘winners and losers’ approach. I think it’s a matter of developing good policy, and we’re going to have to see how this plays out.”

But AI safety hardliners believe they now have the momentum on their side — and they aren’t planning to let up.

“I strongly believe we’re heading towards mandatory,” said Bannon. “Not simply testing, but actual regulatory, some sort of atomic energy-type regulatory approval of this.”

“Right now, the Office of Technology in the White House will only respond when you put pressure on it,” Bannon added. “We intend to ramp up the pressure campaign.”