It is Anthropic's understanding that the government believes there is a method of bypassing, or "jailbreaking," a safeguard that would prevent Fable 5 from being used in identifying software vulnerabilities, the company said.
The order comes just as a previous dispute between Trump administration officials and IPO-bound Anthropic showed signs of easing across parts of the U.S. government.
Anthropic's relationship with the government ruptured this year after it refused to allow the U.S. military to use its AI models for domestic surveillance and fully autonomous weapons systems. The government responded by putting Anthropic on a supply chain blacklist, set to take effect ?later in the year.

The action also marks a major escalation of U.S. efforts to halt foreign adversaries' AI capabilities. For years, U.S. export controls have focused on the chips and tools that power AI rather than on restricting foreign access to AI itself.
Anthropic said the government has given it only "verbal evidence of a potential narrow, non-universal jailbreak".
"We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people," the company said.
The government directive and Anthropic's response highlight growing tension between AI developers and regulators over how to assess risks from so-called "jailbreaks," or methods used to bypass model safeguards.
As recently as Wednesday, Anthropic had called for greater U.S. oversight of AI, including the ability to block models with unacceptable risks. It said, however, the government action on Friday did not follow principles of fair and fact-based regulation.

The Pentagon's chief information officer, Kirsten Davies, said in a post on X that the Defense Department supported prioritizing national security.
"Some things are simply more important than revenue cycles, clickbait, and pre-IPO valuation. America First. Always," Davies said.
Anthropic confidentially filed for a U.S. IPO last month, edging ahead of rival OpenAI in the race to reach public markets.
SOPHISTICATED CYBERATTACKS
Earlier this week, Anthropic rolled out an AI model named ?Claude Fable 5, representing a new tier of capability it calls "Mythos-class." The model is accompanied by guardrails barring its use in risky areas such as cybersecurity, which some users have complained are "overly broad," Anthropic said.
Experts have said that Mythos models, in the wrong hands, could dramatically accelerate sophisticated cyberattacks, particularly in sectors such as banking that rely on complex, interconnected, and often decades-old technology systems.
Anthropic said it had worked with the U.S. government, among others, on safety ahead of the Fable launch and that models from rival AI providers showed a similar ability to unearth minor bugs in code.
"The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all ?our customers to ensure compliance. Access to all other Anthropic models will not be affected," Anthropic said.
Anthropic said that it believed there was a "misunderstanding" and that it is working to restore access to the models as soon as possible.
"If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers," the company said.
Amazon's cloud unit AWS said late on Friday that Anthropic has asked it to revoke access ?to the models for "all users in all regions."
A U.S. official confirmed that the Commerce Department had issued an export control directive to suspend all access to Fable 5 and Mythos 5 by foreign nationals.
Dean Ball, a former White House official who contributed to the AI Action Plan the administration issued in the summer of 2025, said in a post on X that the order suggests all "non-Americans" would be restricted from using ?Anthropic's latest models, including those based in the U.S.
"This means you should expect to have to prove your citizenship to use Anthropic models," Ball said.
Several key Anthropic personnel, including co-founder Chris Olah, AI researcher Andrej Karpathy and philosopher Amanda Askell, were born outside the United States. Reuters was unable to determine their citizenship status, and an Anthropic spokesperson declined to comment on whether such staff would lose AI model access.

Reporting by Ananya Palyekar in Bengaluru, Mrinmay Dey and Chris Thomas in Mexico City, Jeffrey Dastin in San Francisco, David Shepardson; Editing by Edmund Klamann and Tom Hogue
Our Standards: The Thomson Reuters Trust Principles.