Escalating tensions over Fable 5 and Mythos 5 AI models explained
Dustin Volz, Julian E Barnes And Ana Swanson
The conflict between Anthropic and the Trump administration escalated again on 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.
Late Friday, Anthropic announced in a blog post that it had disabled all customers’ access to its most advanced artificial intelligence systems, known as Fable 5 and Mythos 5, after receiving an instruction from management to suspend access to any foreign nationals.
The move shocked former US officials and cybersecurity experts; Many of them questioned the validity of the action and said it deviated from the hands-off approach to policing the burgeoning artificial intelligence industry that President Donald Trump endorsed earlier this month.
Anthropic said the directive did not disclose the national security concerns that triggered it. But the company added that the government had become aware of the method of “jailbreaking” or bypassing security restrictions in Fable 5, which it said was intended to limit the customer’s ability to misuse the product for hacking or other potential harm. Anthropic countered this, saying fears about the jailbreak method were exaggerated.
The announcement increased tensions between the Trump administration and Anthropic, which secretly filed for an initial public offering after a financing round valued at nearly $1 trillion earlier this month.
The two sides argued for months over how Anthropic’s AI systems could be used in military and intelligence environments, resulting in the Pentagon labeling the company as a supply chain risk.
On Saturday, senior Trump administration officials and Trump allies stepped up their attacks on Anthropic.
“Three months ago @DeptofWar kicked @AnthropicAI out of our building forever,” Secretary of Defense Pete Hegseth said on social platform X. “Every day he proves why this was the right move.”
David Sacks, a venture capitalist who until recently worked as the executive AI czar, accused Anthropic of being reckless in the launch of its latest model, called Fable 5, in a lengthy social media post on Saturday.
Sacks, who said he spoke to many people inside and outside the administration about the directive, said the administration asked Anthropic’s CEO, Dario Amodei, to address the jailbreak issue after Anthropic and a “highly reliable trusted partner” of the government came forward with the research.
“Dario refused,” Sacks said.
A person briefed on Amodei’s conversations disputed that account, saying Anthropic was happy to discuss concerns.
Sacks did not name the “trusted partner” he mentioned in his post. However, according to sources close to the matter, many technology companies, including Amazon, met with the White House about security issues. Many said Amazon CEO Andy Jassy’s message detailing security issues was the most effective, raising concerns about the capabilities of the new Anthropic model.
But many of those officials said a separate document from Amazon describing security concerns about Anthropic’s model was misleading. The relevant capabilities that the document highlights with the Anthropic model are also available in OpenAI’s top model, 5.5.
An Amazon spokesman declined to comment on the exact nature of the discussions with the White House. “It is not uncommon for governments to seek our advice on potential security risks,” the spokesman said. “We do not share details of these discussions when such incidents occur.”
Administration officials called Anthropic officials at 1:15 p.m. Friday and gave them 90 minutes to remove their most advanced model, citing an unidentified national security concern, according to people with knowledge of the discussions. Anthropic officials requested more information and worked to find out exactly what the concern was, as the Department of Commerce’s review and testing of Fable did not reveal significant concerns.
Then at 5:21 p.m., Anthropic was notified that the Trump administration had imposed export controls that effectively forced the company to withdraw its model, which consumers had just begun to access.
Discussions on resolving the dispute continue. Commerce Secretary Howard Lutnick spoke with Anthropic officials on Friday and was scheduled to hold another meeting with top company officials Saturday evening, according to people familiar with the discussions and plans.
Some experts said the Trump administration either misunderstood or deliberately misinterpreted what was happening. Katie Moussouris, CEO of Luta Security, said she saw the investigative article on social media that triggered the administration’s actions.
“This is not a jailbreak,” he said, but rather a defensive maneuver designed to limit misuse of a model. If the target is national defense, this is an own goal.
Earlier this month, the Trump administration issued an executive order asking tech companies to allow the government to voluntarily review their new models before making them available to the public. But it did not give the government a formal role in approving the release of these people.
The order had been delayed following fierce debate in which AI companies and some sympathetic government officials opposed government interference in model distribution. The dispute comes on the heels of Anthropic’s very limited release of Mythos, an AI model that has been very successful at discovering and weaponizing new cybersecurity vulnerabilities, prompting 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 consider the long-term consequences of such a move.
But the measure may be limited to Anthropic, which has been subject to multiple attacks from management in recent months.
In February, amid the Pentagon’s conflict with Anthropic, Trump called the company a “radical left, woke company” and “leftist lunatics” trying to dictate how the government wins and fights wars.
“I am directing ALL Federal Agencies in the United States Government to IMMEDIATELY STOP all use of Anthropic technology,” the President wrote on social media. “We don’t need it, we don’t want it and we won’t do business with them again!”
U.S. officials say some administration officials have been seeking a way out of the dispute in recent weeks. White House and intelligence officials have touted a secret contract between Anthropic and the National Security Agency that 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.
NSA officials responsible for digital interception and government cybersecurity were not involved in Friday’s decision, according to people familiar with the matter.
Many U.S. officials have said the new technology developed by Anthropic is too important to national security to let the dispute with the Pentagon hinder cooperation. But officials higher up in the Pentagon remain upset with the company and insist that it sign the same contractual terms that other AI companies have adopted.
In a letter dated June 12 reviewed by New York TimesLutnick told Amodei that a special license would be required for the company to distribute the Mythos and Fable 5 models “to all destinations worldwide” and share them with non-U.S. citizens.
“Failure to comply will result in immediate criminal and civil penalties as provided by law,” Lutnick wrote.
The government has controlled AI models before, but past restrictions were more targeted. A measure introduced by the Biden administration early last year placed restrictions on companies’ ability to assign so-called weights—specific numerical values that tell the model how much importance it should give to different pieces of data—for certain AI models.
The Trump administration’s action goes far beyond that by prohibiting Anthropic from sharing the model with any foreign country or any foreign national without a license.
Restrictions on participation by foreign nationals, even green card holders, can be particularly chilling in an industry that relies on foreign talent. In the past, government officials generally reserved such restrictions for the most sensitive technologies, such as weapons systems.
Chris McGuire, a former Biden administration technology official who is now a senior fellow at the Council on Foreign Relations, said the administration’s current policy is that AI chips can be exported to China, but Canadian green card holders in the U.S. will not be able to access leading U.S. models.
“This is ridiculous,” he added.
This article was first published on: New York Times.
The Business Briefing newsletter delivers big stories, exclusive news and expert insights. Sign up to receive it every weekday morning.
