Anthropic has disabled access to new AI models, following US export restrictions

Anthropic was forced by the US government on Friday, three days after launching two new powerful AI models, to restrict access to these models. The Trump administration has issued an export ban on the models, called Fable 5 and Mythos 5, on national security grounds.

The measure applies to “all foreign nationals, inside or outside the United States,” according to a letter the AI ​​company received on Friday. The government decision therefore also applies to Anthropic’s foreign employees, the company writes a statement on the website. In order to comply with the decision, the company therefore had to ‘immediately disable’ access to Fable 5 and Mythos 5 for all customers.

Last Tuesday, Anthropic made the Fable 5 model available to Claude users. This latest, powerful model delivers according to Anthropic “exceptional achievements” in software development, image processing, scientific research “and many other areas.” Since Friday, Claude users can no longer access the projects they started with Fable 5. Mythos 5 was accessible to a small group of cyber security professionals and infrastructure providers since Tuesday. This model is particularly good at finding vulnerabilities in software.

According to Anthropic, the Trump administration has not specifically explained the “reasons for national security concerns.”

News site Axios, who has the government letter in his handswrites that the measure may remain in force for several weeks. According to the site, Commerce Secretary Howard Lutnick made the decision after the department was alerted by another company that it could have “jailbroken” the Anthropic models, thereby bypassing security measures designed to prevent malicious use. The restrictions would apply until the U.S. security apparatus is strengthened and better prepared for potential cybersecurity risks.

Anthropic says that the security measures are “so strong” and widely implemented that “many users have complained about them.” Ahead of Fable’s launch, the company says it worked with the US government and the British state agency AI Security Institute (AISI) and tested the model’s security measures for “thousands of hours”.

“We dispute that the discovery of a possible circumvention is grounds for recalling a commercial model that has been deployed on hundreds of millions of people,” writes Anthropic, which considers the case “a misunderstanding.” The company warns of the impact of the US government’s ban on the entire AI sector: “If this standard were to be applied across the industry, it would effectively halt all new implementations of advanced AI models,” the company said.

Give a gift





Part





Email the editor