The White House’s last-minute restrictions on Anthropic’s new artificial intelligence model are raising fundamental doubts about President Donald Trump’s 2-week-old effort to place guardrails on the technology.
And they’re undercutting the president’s promise of a light-touch approach to regulating advanced AI, according to tech industry representatives, policy experts and at least one administration official.
The June 2 order was meant to set up a process that could address the technology’s cybersecurity risk without unduly disrupting the U.S. tech industry, allowing companies to keep innovating and competing with China for supremacy in AI development. Instead, unspecified security concerns about Anthropic’s newest products ignited an 11th-hour scramble by the administration last week, leading to a federal edict that prompted the company to abruptly cut off access to its new Fable model for all users Friday night.
“It’s clear that there has been a vibe shift in the Trump administration on AI, and the priorities have somewhat apparently shifted to first and foremost taking down Anthropic,” said Adam Thierer, a senior technology and innovation fellow at the libertarian R Street Institute.
The situation in some ways mirrors this year’s earlier standoff between Anthropic and the Defense Department, which culminated with the Pentagon labeling the company a supply chain risk to national security. Industry advocates warned at the time that the fight would hobble America in the global race to develop advanced AI. But relations between Anthropic and the government soon thawed somewhat — driven in part by the government’s eagerness to gain access to Mythos, a powerful new Anthropic model whose public-facing version was released last week as Fable 5.
Now, in response to the government slapping restrictions on Anthropic’s new models that bars non-citizens from using them, Anthropic has cut off access to both Fable 5 and Mythos 5, its newest, most tightly-controlled models with elite cyber capabilities.
Asked about whether the new export control directive undermined the president’s recent AI executive order, White House spokesperson Kush Desai said the Trump administration “is collaborating with AI industry leaders to balance cutting-edge innovation with national security concerns that affect both the United States and our allies.”
“The United States is by far the world leader in the global AI race, and President Trump is committed to ensuring America’s technological dominance,” said Desai.
A spokesperson for Anthropic declined to comment.
After weeks of back-and-forth debate, Trump’s executive order laid out a framework through which top tech companies would voluntarily submit new AI systems to be vetted by the government 30 days before their planned release.
The plan explicitly rejected earlier proposals that would have made the vetting process mandatory or forced companies to get a license from the government before releasing their models. The final contours of the order were shaped in part by venture capitalist and former White House AI czar David Sacks, who still holds the president’s ear.
But on Friday, the White House moved to limit access to Anthropic’s Fable after Amazon CEO Andy Jassy raised concerns that users could bypass the model’s guardrails, according to two administration officials and a White House official granted anonymity because they were not authorized to discuss the episode.
Anthropic said the move forced it to disable the product with virtually no notice — triggering waves of anxiety across the AI sector, as lobbyists and industry advocates struggled to square the administration’s action with the order’s promise of an approach free of overbearing regulation.
“We don’t want to have a situation where politically unfavored actors or their models are all of a sudden finding themselves late on a Friday afternoon, having to pull models off the global market to satisfy the demands of certain people in an administration,” Thierer said. He added that the executive order “was supposed to bring some order to the situation, but it’s clear that it really has not done that.”
Like several people interviewed for this article, Thierer said it was now clear that the government’s proposal to vet advanced AI models was “not voluntary at all.” And others said the forced takedown of Fable strongly suggests the government will require new AI systems to be licensed by the government — even if the executive order says otherwise.
“We’re concerned that this resembles something of an ad hoc licensing regime, which we think would be antithetical to promoting U.S. technology around the world,” said Paul Lekas, head of global public policy and government affairs at the Software & Information Industry Association, a tech trade group that includes Anthropic and Amazon among its members.
“Regardless of statements the administration may make about this applying only to one developer, it definitely raises concerns among the broader industry that they could be subject to a similar action,” Lekas said. The lobbyist added that the Friday action “represents the kind of approach that industry has long been opposed to — and the administration, too, has been opposed to.”
Some AI experts said the history of bad blood between Anthropic and some administration officials added a disturbing element to the security debate.
“Anytime you move from a standards-based or law-based approach — where at least you know the hoops you have to jump through — to a space where your connections and the decisionmakers’ opinions of you and your company are the primary drivers, that just politicizes everything in an industry,” said Neil Chilson, head of AI policy at the Koch-backed Abundance Institute.
Chilson added that he saw the administration’s Friday action against Anthropic as “less of a licensing regime and more of a beauty contest.”
“It’s more like, ‘We know it when we see it whether or not we like it,’” said Chilson. “And that is much more uncertain than a licensing regime.”
Even one Trump administration official — granted anonymity to freely discuss the impact of the White House action against Anthropic — suggested the export control will continue to undermine the order for as long as it remains in place.
“If this blows over, if the restrictions are lifted tomorrow, then I would say, ‘Oh OK, this is still kind of consistent with the [executive order], and we’ll just keep going for it,” said the official. “If the situation is still fucked a week from now, then I think we have a clear understanding of what’s going on.”
The official added that a drawn-out fight over the export control would be “a huge problem” for the American AI industry.
“It means that every model going forward needs to ask [the] government’s permission for whether it can be released,” the official said. “That’s an extremely bad situation, and it would completely cripple the whole industry.”
The government’s export control blocks foreign nationals from accessing Anthropic’s most advanced models. One senior official at a top AI company, granted anonymity to freely discuss the White House action, worried that if the control persists or restrictions spread to other firms, it would cause researchers from foreign countries to look outside of the U.S. for work.
“It would really put the whole AI industry at a disadvantage and be a recruiting and retention problem for AI researchers,” said the industry official.
It’s not yet clear when or whether the latest standoff between Anthropic and the White House will be resolved. Meetings on Monday between the two sides produced no immediate announcements of a truce, with one White House official telling POLITICO the impasse could persist for more than a few days.
The administration official who spoke to POLITICO said that “the vast majority of people that I talked to in the administration think this path is a terrible idea.”
“It’s a small number of people … who are now taking the lead on this,” said the administration official.
Chris McGuire, a senior fellow for China and emerging technologies at the Council on Foreign Relations and a former Biden administration official, said the White House now faces two choices: make up with Anthropic and bring Fable back online, or be forced to apply these same types of controls broadly to other advanced AI models.
“I kind of think that they’ll probably do the former, because if they did the latter, it would be devastating to AI competitiveness,” he said.
Dana Nickel contributed to this report.
