Amazon Warning Prompted U.S. Order to Disable Anthropic Models

U.S. officials ordered Anthropic to disable its Fable 5 and Mythos 5 models after internal reports found Fable 5 could be prompted to produce information tied to cyberattacks and software vulnerabilities.

Anthropic researchers reported that specific queries caused Fable 5 to output details exposing flaws in at least four software programs. Amazon, a major backer of Anthropic with billions invested, raised the concerns directly with senior Trump administration officials. Security researchers outside the companies began to reproduce and test the findings.

Officials told Anthropic to either fix the identified flaws or remove the model from service. After discussions involving National Cyber Director Sean Cairncross and Commerce Secretary Howard Lutnick, the administration directed the Commerce Department to bar foreign governments, companies and individuals from accessing the models. The Commerce Department issued the order with President Trump’s approval; a senior White House official expressed concern the restriction could slow innovation.

Anthropic disabled Fable 5 and Mythos 5 for the affected users and described the interruption as temporary while it works to address the issues and restore access. The company apologized to customers for the disruption.

Anthropic disputed the severity of the risk, saying the flagged vulnerabilities were basic and could be found by other models without special bypasses. The company wrote in a public response: “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. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”

The shutdown came while Anthropic prepares for a possible initial public offering; extended restrictions or outages could affect user retention. Security teams inside and outside government continued to stress-test the models to determine how easily the reported vulnerabilities could be reproduced and exploited.

Amazon is one of Anthropic’s largest investors. The Commerce Department enforces export controls on critical technology and used the ban on foreign users as a mechanism while technical fixes are developed and validated.