Anthropic’s sudden disabling of its latest AI models for customers has sent frustration rippling across the cyber and technology sectors.
The tech giant paused access to Mythos 5 and Fable 5 for all users in response to a June 12 White House directive to ban the use of the models by any foreign nationals, both inside and outside the US.
The move comes as frontier AI models are poised to alter how companies spot and fix software vulnerabilities and defend against AI-powered threats—and as Anthropic was looking to allow more businesses beyond an initial set to access the technology. The White House’s and Anthropic’s decisions muddy an already murky discussion around how models should be governed and raise questions about whether users should go all in on models before they are fully vetted.
The halt will cut short an early window of cybersecurity experimentation, said Camille Stewart Gloster, the CEO of CAS Strategies LLC, a cybersecurity services advisory firm.
“Mythos was a limited preview, and many companies with access were using it to find vulnerabilities, test defensive workflows, and understand what frontier models could do in security contexts,” said Gloster, who previously worked as a deputy national cyber director in the Biden White House. “Security leaders have been warning that there is a closing window to build resilience before attackers fully operationalize frontier AI. This kind of restriction narrows that window further.”
Restricting Access
In a letter sent Sunday to Commerce Secretary Howard Lutnick and National Cyber Director Sean Cairncross, more than a hundred cyber and tech executives from companies including
The tech leaders said it was “essential” to give AI to coders and security teams so they can find and fix flaws in their systems.
“This action has taken the best models away from defenders, created market uncertainty, and risked America’s AI leadership without any real risk to justify it,” the letter said.
Anthropic’s pause on its latest models comes as the AI giant was working to expand access. Earlier this month, Anthropic opened up Mythos to more than 150 organizations.
Josh Kamdjou, founder and CEO of email security platform Sublime Security, said access to the Mythos model shouldn’t necessarily be restricted right now, but there may come a time when models have more capability to do serious harm to US companies.
“There is a big difference between understanding of vulnerability and actually weaponizing it, and conducting an operation and achieving an objective. There’s a huge delta between that,” Kamdjou said. He added, “We’re going to continue to see messiness because of how quickly these models are changing and improving and the perception of their capability. And frankly, we don’t know how to exactly regulate and what boundaries and things like that.”
The government’s order to suspend access to foreign nationals would also have been challenging for companies.
“Technology and cybersecurity companies are powered by global talent. If restrictions are too broad, they can create friction inside teams that depend on non-U.S. citizens in engineering, research, security, and product roles,” Gloster said. “A blunt restriction can hurt defenders, slow patching, fracture teams, and push workarounds into less transparent channels.”
Path Forward
The face off between Anthropic and the White House may ultimately hinder future collaboration between the public and private sectors, a goal the administration had laid out in its AI executive order signed in early June.
“I don’t think the chilling effect is on the companies that want access to the models. I think the chilling effect is on the models themselves,” said a cyber legal expert, who spoke anonymously due to the nature of their work.
Voluntary collaboration between AI giants and the government is “what might be at risk because if Anthropic, in fact, believes that the government’s testing is producing inaccurate results and still taking such aggressive action that has a huge impact, then why would a future frontier model want to do anything to collaborate with the US government?” the cyber legal expert said.
The pause is the latest fallout from a regulatory gap around what companies need to do to vet their sophisticated models with the government.
“Companies will have to perform for decision makers and hope they make the cut,” said Neil Chilson, head of AI Policy at the Abundance Institute, a public policy nonprofit dealing with technological innovation.
“Instead of competing in the market by trying to build new and powerful tools to keep their customers happy, companies will spend their time and resources competing for political favors by trying to keep their regulators happy,” Chilson, a former acting chief technologist at the Federal Trade Commission, said in an email.
Despite the uncertainty, many companies stand ready to experiment with the models and leverage AI to spot vulnerabilities.
“The companies who are thinking smartly about this are going to want to get access to as much capability and information as they can,” said Jordan Rae Kelly, head of cybersecurity for the Americas at FTI Consulting and former White House cyber official during the first Trump administration.
“There will probably be some frustration that we seem to have slowed this down,” she added.