Anthropic asked for regulation. Washington went much further

Anthropic’s longstanding push for regulation is coming back to bite it.

The artificial intelligence company, now valued at close to $1 trillion, has spent years touting its dedication to safety. But for the second time this year, Anthropic has found itself caught in the Trump administration’s crosshairs, this time out of concern for the safety of its newest models.

Late Friday, a couple hours after SpaceX wrapped up its first day of trading following a record IPO, Anthropic said it received an export control directive, ordering the company to suspend access to its latest Claude models, Fable 5 and Mythos 5, “by any foreign national, whether inside or outside the United States, including foreign national Anthropic employees.”

The administration cited “national security authorities” but didn’t specify its concern, Anthropic said. The directive landed just days after Anthropic CEO Dario Amodei published an essay advocating for more “serious and binding regulation of AI,” including the ability to block models if they are deemed unsafe. 

“Frontier AI models, like airplanes, should be required to go through technical testing and auditing, and their release should be blocked or reversed as a threat to public safety if they do not meet high standards of safety,” Amodei wrote. 

It was the latest public statement from Anthropic encouraging greater government oversight of the rapidly evolving AI industry. Since splitting from OpenAI to start Anthropic in 2021, Amodei and his top executives have been staunch advocates for AI regulation and have supported legislation at both state and federal levels.

The company lauded an AI executive order that President Donald Trump signed earlier this month as an “important step.”

But Friday’s action wasn’t the kind of oversight Anthropic had in mind. The company said in response that it disagreed that the Trump administration’s finding was cause for a recall, and it characterized the disruption as a “misunderstanding.”

“As we have stated publicly, we believe the government should have the ability to block unsafe deployments, as part of a statutory process that is transparent, fair, clear, and grounded in technical facts,” Anthropic said in a blog post. “This action does not adhere to those principles.”

Senior Anthropic employees flew to Washington, D.C., to meet with members of the Trump administration on Monday. The company told CNBC that “both parties are working quickly to get this resolved.”

Anthropic hasn’t said when it expects its models to come back online.

‘This sure looks mandatory’

Trump’s executive order, 10 days ahead of the directive against Anthropic, was thin on specific details.

It asked companies, on a voluntary basis, to provide models to the federal government to assess their capabilities ahead of a full release. The order also gave administration officials 60 days to develop and maintain the relevant review frameworks for AI companies to consider.

There’s no indication that the forced suspension of Anthropic’s latest models was tied to the executive order. Rather, according to The Wall Street Journal, the move was prompted by conversations that Amazon CEO Andy Jassy had with U.S. officials, including Treasury Secretary Scott Bessent. Citing people familiar with the matter, the Journal said Jassy told the administration that Amazon researchers used a series of prompts to get Fable 5 to serve up information that could aid cyberattacks.

While the president’s executive order called for voluntary adherence, Friday’s directive had a very different tone, said Daniel Remler, a senior fellow at the Center for a New American Security.

“This sure looks mandatory if there are going to be consequences for not doing what the government says,” Remler told CNBC in an interview.

Fable 5 and Mythos 5 built on the April release of Claude Mythos Preview, a powerful offering that excels at identifying security vulnerabilities within software. Anthropic limited the rollout to a select group of companies as part of a cybersecurity initiative called Project Glasswing.

Mythos 5 was still limited to a select group of users, but Anthropic made Fable 5 available to its enterprise customers and paid subscribers. The company said the broad release was possible because of new safeguards that block responses in specific high-risk areas, including cybersecurity and biology.

Anthropic worked with government agencies to test the models ahead of the release and received approval to deploy them, according to a person familiar with the discussions who asked not to be named in order to discuss confidential matters. There was no communication from the government of a national security threat, the person said.

“To go from an environment in which we had no controls on model access at all to basically using this tool to take down one company’s model in a few short hours is pretty remarkable,” Remler said.

He added that, given the volatile policy environment, it’s the kind of “reactive regulation” the company should probably have anticipated. 

“It’s hard to manage risks with a technology that’s changing so quickly all the time,” he said.

‘Ball is in Anthropic’s court’

It’s all happening at a pivotal moment for Anthropic. 

The company and its chief rival, OpenAI, both confidentially filed their IPO prospectuses recently, setting up potentially historic share sales for investors eager to jump into AI. SpaceX’s three-day rally following its IPO is seen by many as evidence of Wall Street’s enthusiasm for new and big opportunities.

Amodei has some loud voices on his side.

Alex Stamos, chief product officer at Corridor and the former security chief at Facebook, penned an open letter on “transparent AI cyber protections,” addressed to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross. The letter, signed by more than 150 executives and technical leaders, asks for the export control directive to be lifted. 

Stamos told CNBC in an interview that rules need to be “based upon science.”

“Those rules need to be written down and transparent. That has not happened,” Stamos said. “There’s nothing that Anthropic or anybody else can look at to say what can and can’t I do.”

He added that for every American company, “the fear is that at any moment you can now, if you run afoul of the administration, be shut down for a completely arbitrary decision capriciously.” 

For Anthropic, clashes with the government have become a big part of the company’s story. Earlier this year, it engaged in a high-profile dispute with the Department of Defense. The DOD declared Anthropic a supply chain risk in March, a designation that requires defense contractors to certify that they will not use Claude models in their work with the military.

In a post on X on Saturday, Defense Secretary Pete Hegseth addressed the government’s latest directive, writing that “every passing day” proves why blacklisting Anthropic was “the right move.”

Anthropic sued the Trump administration in an effort to reverse the supply chain risk designation. That litigation is ongoing.

Stamos said the administration clearly has a problem with Anthropic, one of the few notable companies to push back on its policies.

“If they were fairly enforcing these rules, they would have to enforce them against OpenAI and Google as well,” he said. 

One vocal critic of Anthropic has been David Sacks, a venture capitalist who formerly served as Trump’s AI and crypto czar. After a company executive published an essay in October on risks of AI, Sacks accused Anthropic in a post on X of “running a sophisticated regulatory capture strategy based on fear-mongering.”

Sacks weighed in after Anthropic received the export control directive, writing on X that the “minimizing language” in the company’s blog post is not in line with its brand as “the AI safety company.”

“The Admin values Anthropic’s technical capabilities and feels that this issue, while serious, should be easily resolved,” Sacks wrote. “The ball is in Anthropic’s court.”

—CNBC’s Kate Rooney contributed to this report

WATCH: Anthropic races to deal with government directive to ban foreign nationals from using AI models