Anthropic pulls back new AI model after U.S. security alert

In a sudden move, Anthropic announced it has suspended its flagship AI tool Claude Fable 5, following a formal request from U.S. national‑security officials concerned about potential security risks. The company said the order required it to halt use of the model by foreign nationals and to disable both Fable 5 and the closely related Mythos 5 for all customers.

The statement, posted on Anthropic’s website, described the tool as too powerful for a general roll‑out. The company explained that it had identified a “jailbreaking” technique that could bypass built‑in safety controls, exposing minor vulnerabilities that, according to the company, could be exploited without the model’s safeguards.

Anthropic’s regulatory tit-for-tat is part of an ongoing legal standoff with the Trump administration over a broader directive that barred U.S. government agencies from using the company’s AI services. The Pentagon’s attempt to enforce the ban was quashed by a U.S. judge, allowing certain agencies to continue purchasing the technology while the lawsuit proceeds.

Earlier in the year, Anthropic had issued a limited preview of Fable 5 to a handful of financial, technology and government leaders to test and patch vulnerabilities before a public release. Critics argued that the company’s claim that the tool was “too powerful” was largely hype, but the latest pause signals increasing concerns from regulators about the potential for generative AI to be weaponised or compromise sensitive systems.

The incident also highlights a new designation by the U.S. Department of Commerce: it called the tool a “supply chain risk,” the first time an American company has received such a label for security reasons. The label is usually reserved for firms based in adversarial nations, underscoring the seriousness of the U.S. government's stance.

While the court ruling keeps Anthropic’s services available to some government bodies, the company’s pause has sparked debate about how the rapid development of powerful AI models should be managed to ensure public safety and trust. As the industry watches, the debate over regulation versus innovation continues to intensify.