Amazon research helped trigger Anthropic export controls on Fable 5

Amazon’s security research helped push a White House export-control decision that knocked Anthropic’s newest models offline for users around the world, exposing how much influence major cloud and AI rivals now have over national restrictions on frontier systems. The episode centers on Fable 5 and Mythos 5, models Anthropic had just launched with new safeguards before the government order landed at 5:21 p.m. ET on Friday.

Anthropic said the directive required it to suspend access to the models for any foreign national, whether inside or outside the United States, including foreign national employees. The company said it had no practical way to separate eligible from ineligible users in real time, so it disabled access for all customers globally. Amazon Web Services later said Anthropic asked it to revoke access to Claude Fable 5 and Claude Mythos 5 for all users to comply with the export-control directive, even though Fable 5 had only recently been made generally available on Amazon Bedrock.

The policy turn followed security concerns raised inside Washington and in the private sector. The Wall Street Journal reported that Amazon researchers, through a series of prompts, got Anthropic’s Claude Fable 5 to reveal information that could be used in cyberattacks. Reuters reported that Amazon chief executive Andy Jassy was among tech leaders who raised those concerns this week with senior Trump administration officials, including Treasury Secretary Scott Bessent. The government later decided that a safeguard around Fable 5 could be bypassed, or “jailbroken,” to identify software vulnerabilities.

Anthropic said the government did not spell out the national-security concern. That silence has sharpened the policy debate around who should shape AI export rules, what evidence should be required before restrictions are imposed, and how quickly such decisions can ripple through the market. Critics said the move swept too broadly. Jimmy Goodrich told Reuters the action was “not well thought-out” because it affected allies as well as adversaries and would bar Canadians and Britons working at Anthropic from doing research and development.

The case has broader competitive consequences for the AI industry. Anthropic said the capabilities at issue were already available in other publicly accessible models, suggesting the restriction may shift demand rather than eliminate risk. It also lands against a backdrop of earlier tensions between Anthropic and the Trump administration over AI policy, and a separate clash with the Department of Defense that had previously declared Anthropic a supply chain risk after negotiations collapsed.