The White House imposed such restrictions “partly over suspicions that a China-linked group had accessed” the Mythos 5 model, a person familiar with the matter told Semafor Saturday. If confirmed, the report would mark a significant development over the broader debate about national security risks associated with Chinese access to American A.I. models.
Concerns about an alleged method to “jailbreak” the models’ security features were reported Friday as prompting the administration to engage in conversations with CEO Dario Amodei, which ultimately led to the models being blocked.
White House AI czar David Sacks said in an X post Saturday that Amodei refused when the Trump administration asked Anthropic to “fix the jail-break or de-deploy the model,” and the restriction was put in “reluctantly.”
“In the past, Anthropic has always said that safety must be top priority and taken super seriously,” he said. “In this case, Anthropic prioritized the continued offering of the consumer model over safety.”
The export restrictions put in place by the Trump administration prohibited access by customers outside the United States, as well as foreign nationals within the U.S. Anthropic responded to Commerce Secretary Howard Lutnick’s directive by “removing access to Fable 5 and Mythos 5 for all users,” citing the need to ensure compliance.
Mythos drew national attention in early April when Anthropic released the model on a limited basis through Project Glasswing, granting access only to a select group of companies “in an effort to secure the world’s most critical software.” The company cited concerns that the model’s advanced cybersecurity capabilities could be misused if released widely.
Worries about a public release came to fruition after researchers at Amazon jailbroke the Mythos 5 model, according to the Wall Street Journal. Chief Executive Andy Jass reportedly told U.S. officials, including Treasury Secretary Scott Bessent, about the revelation, setting off a firestorm Friday within the White House involving chief of staff Susie Wiles and President Donald Trump.
Katie Moussouris, CEO of Luta Security, reviewed the report shared by Anthropic. She told the Washington Examiner that Amazon researchers used prompts to “generate output that could be used to find vulnerabilities and generate test scripts that could be turned into proof of concept code.”
“As a leading cloud provider that serves a large number of private and public sector customers, it’s not uncommon for governments to seek our counsel on potential security risks,” an Amazon spokesperson told the Washington Examiner. “When they occur, we don’t share details of these discussions.”
Anthropic also said in the Friday statement it disagrees with the administration’s decision “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 were applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
AMAZON RESEARCHERS USE ANTHROPIC AI TO FIND CYBERSECURITY WEAKNESSES
Secretary of War Pete Hegseth brought up the Pentagon’s prior standoff with Anthropic in an X post Saturday night, saying “every passing day proves” it was the right move to end the government’s contract with the company.
Anthropic filed to go public earlier this month. The two models were released publicly on Tuesday.
