Trump reignites Anthropic feud after takedown of latest AI model

The feud between Anthropic and the Trump administration escalated again Saturday after the US government took the unusual step a day earlier of demanding that the artificial intelligence company cut off foreign access to its latest models, as top officials suggested the dispute was unlikely to resolve quickly.

Late on Friday, Anthropic disclosed in a blog post that it had disabled access for all customers to its most advanced AI systems, known as Fable 5 and Mythos 5, after receiving a directive from the administration to suspend access to any foreign national.

The move shocked former U.S. officials and cybersecurity experts, many of whom questioned the validity of the action and noted that it diverged from the hands-off approach to policing the booming AI industry that President Donald Trump had endorsed earlier this month.

Anthropic said the directive did not explain the national security concerns that prompted it. But the company added that the government had said that it became aware of a method to “jailbreak,” or bypass, security restrictions on Fable 5 intended to limit a customer’s ability to abuse the product for hacking or other potential harms. Anthropic countered that the fears about the jailbreak method were overblown.

The announcement inflamed tensions between the Trump administration and Anthropic, which earlier this month confidentially filed for an initial public offering following a funding round valuing it at nearly $1 trillion.

The two sides have sparred for months over how Anthropic’s AI systems could be used in military and intelligence settings, culminating with the Pentagon labelling the company a supply chain risk.

On Saturday, top Trump administration officials and allies of Trump’s stepped up their attacks on Anthropic.

“Three months ago, @DeptofWar kicked @AnthropicAI out of our building—forever,” Defence Secretary Pete Hegseth said on the social platform X. “Every passing day proves why that was the right move.”

David Sacks, a venture capitalist who until recently worked in the administration as an AI czar, accused Anthropic in a lengthy social media post Saturday of being reckless with the release of its latest model, dubbed Fable 5.

Sacks, who said he had spoken to many people inside and outside the administration about the directive, said the administration had asked Dario Amodei, Anthropic’s CEO, to fix the jailbreak issue after a “highly credible trusted partner” of Anthropic’s and the government’s came forward with research.

“Dario refused,” Sacks said.

A person briefed on Amodei’s conversations disputed the account, saying that Anthropic was happy to discuss the concerns.

Sacks did not name the “trusted partner” he referred to in his post. But multiple technology firms, including Amazon, spoke with the White House about the security issues, according to people familiar with the matter. Several of the people said a message from Amazon’s CEO, Andy Jassy, detailing security issues was the most influential, raising concerns about the capabilities of the new Anthropic model.

But several of these officials said a separate document from Amazon explaining the security concerns with Anthropic’s model was misleading. The concerning capabilities that the document highlighted with Anthropic’s model are also present in OpenAI’s top model, 5.5.

An Amazon spokesperson declined to comment on the exact nature of its discussions with the White House. “It’s not uncommon for governments to seek our counsel on potential security risks,” the spokesperson said. “When they occur, we don’t share the details of these discussions.”

Administration officials called Anthropic officials at 1:15 p.m. Friday and gave them 90 minutes to pull their most advanced models down, citing an undefined national security concern, according to people briefed on the discussions. Anthropic officials asked for more information and worked to learn what the precise concern was, since the Commerce Department’s review and testing of Fable did not reveal significant concerns.

Then, at 5:21 p.m., Anthropic was notified that the Trump administration was imposing export controls that effectively forced the company to pull down its model, which consumers had just begun to have access to.

Discussions about resolving the dispute are continuing. Howard Lutnick, the commerce secretary, spoke to Anthropic officials Friday, and was set to have another session with senior company officials Saturday evening, according to people briefed on the discussions and plans.

Some experts said the Trump administration was either misunderstanding or deliberately misconstruing what transpired. Katie Moussouris, the CEO of Luta Security, said on social media that she had seen the research paper that prompted the administration’s actions.

“It’s not a jailbreak,” she said, but rather a defensive manoeuvre designed to limit the misuse of a model. “If national defence is the goal, this is an own goal.”

Earlier this month, the Trump administration issued an executive order that asked technology companies to voluntarily let the government review their new models before releasing them to the public. But it did not give the government an official role in approving their release.

The order had been delayed after a fierce debate in which AI companies — and some sympathetic government officials — had pushed back against the government interfering in model deployment. The dispute came on the heels of Anthropic’s very limited release of Mythos, an AI model so good at discovering and weaponizing new cybersecurity vulnerabilities that it sparked widespread concerns about its potential to wreak havoc.

The new restrictions on Anthropic have raised questions about the implications for other AI models that may have similar capabilities. Former officials and technology experts said the administration did not appear to have thought through the longer-term ramifications of such a move.

But the measure could be limited to Anthropic, which received a fusillade of attacks from the administration in recent months.

In February, amid the Pentagon’s clash with Anthropic, Trump called the company a “radical left, woke company” and “leftwing nut jobs” working to dictate how the government wins and fights wars.

“I am directing EVERY Federal Agency in the United States Government to IMMEDIATELY CEASE all use of Anthropic’s technology,” the president wrote on social media. “We don’t need it, we don’t want it, and will not do business with them again!”

Some administration officials have in recent weeks been looking for an off-ramp in the dispute, U.S. officials say. White House and intelligence officials have pushed forward a classified contract between Anthropic and the National Security Agency, which would allow the spy agency to use the company’s technology for a variety of purposes, including intelligence analysis and detecting new computer vulnerabilities.

The government has controlled AI models before, but past restrictions were more targeted.

Officials at the NSA, which is responsible for digital eavesdropping and government cybersecurity, were not involved in Friday’s decision, according to people familiar with the matter.

Many U.S. officials said the new technology Anthropic had developed was too important to national security to allow the dispute with the Pentagon to block cooperation. But inside the top ranks of the Pentagon, officials remain upset with the company, insisting that the firm sign on to the same contractual provisions that other AI companies have embraced.

In a June 12 letter viewed by The New York Times, Lutnick told Amodei that a special license would be required for the company to distribute its Mythos and Fable 5 models “to all destinations worldwide,” as well as to share them with non-U.S. citizens.

“Failure to comply will result in prompt criminal and civil penalties, as provided for by law,” Lutnick wrote.

The government has controlled AI models before, but past restrictions were more targeted. A measure introduced early last year by the Biden administration placed restrictions on companies’ ability to export the so-called weights for specific AI models – the proprietary numerical values that tell the model how much importance to place on different pieces of data.

The measure taken by the Trump administration goes far beyond that, by barring Anthropic from sharing the model with any foreign country or any foreign national without first obtaining a license.

The restrictions on the involvement of foreign nationals, even those with green cards, could be particularly chilling in an industry that relies on foreign talent. In the past, government officials have often reserved that type of restriction for the most sensitive technologies, like weapons systems.

Chris McGuire, a former Biden administration technology official who is now a senior fellow at the Council on Foreign Relations, said that the administration’s current policy is that AI chips can be exported to China, but Canadian green card holders in the United States cannot access leading U.S. models.

“That’s absurd,” he added.

This article originally appeared in The New York Times.

The Business Briefing newsletter delivers major stories, exclusive coverage and expert opinion. Sign up to get it every weekday morning.