Anthropic cuts access to AI models after US order

Anthropic PBC suspended access to its AI models Fable 5 and Mythos 5 on Friday following a US government directive blocking foreign nationals—including its own employees—from using them. The order, received at 5:21 PM New York time, did not specify the exact security concerns but suggested the government believed a method to bypass Fable 5’s safeguards could aid hacking. The company stated in a blog post that the directive forced it to disable access for all users to comply. Fable 5, launched publicly just three days earlier, is a restricted version of Mythos 5, an advanced AI model withheld from the public due to fears of its ability to identify software vulnerabilities. Mythos 5 has only been released to select companies, including the EU, which previously gained access after negotiations. Anthropic claimed it reviewed the alleged 'jailbreaking' method and found no evidence that Fable 5 posed unique risks compared to other public AI models. The company argued that recalling the model based on a narrow vulnerability could set a precedent halting future AI deployments across the industry. Researchers at Amazon had previously identified vulnerabilities in Anthropic’s models, with Amazon CEO Andy Jassy reportedly raising concerns with US officials before the restrictions were imposed. The EU responded to the development by emphasizing the need for 'technological sovereignty,' noting that the restrictions further highlight Europe’s reliance on independent AI capabilities. Anthropic’s move follows broader US efforts to regulate advanced AI models amid growing concerns over national security and cyber risks.