Anthropic disables Claude Fable 5 and Mythos 5

Anthropic has suspended access to two of its most advanced AI systems after US national security authorities raised concerns about a potential bypass technique, according to the BBC. The company said it was ordered to disable both Claude Fable 5 and Mythos 5 for all customers, a step that abruptly cuts off tools it had described as unusually capable.

The BBC reports that officials have not publicly identified the specific national security issue, but believe they have found a way to “jailbreak” Fable 5 — a method for getting around built-in restrictions. Anthropic said it reviewed a demonstration and found a small number of vulnerabilities it described as previously known and minor, adding that similar issues could be discovered using other publicly available AI models without any special bypass. Even so, the outcome was not a patch, a warning label, or a limited rollback: it was a blanket suspension.

The episode lands in the middle of a widening fight over who gets to use frontier models and under what conditions. Anthropic had run a private release of Fable 5 in April, offering pre-release access to a handful of organisations to probe for weaknesses, the BBC says. That kind of controlled testing is normally sold as a safety valve; here it becomes part of the case for restricting distribution when the tool is deemed powerful enough to be misused.

It also shows how quickly “security” turns into jurisdiction. The European Union had only recently gained access to Mythos after weeks of talks, according to the BBC, and EU officials framed the suspension as evidence for “technological sovereignty” — a polite way of saying that access to critical software can be turned on and off abroad. In London, a professor at Queen Mary University of London warned that restricting access could reduce independent testing and slow collaboration with governments.

In the US, the line between procurement and regulation is blurring. The BBC notes that Anthropic is involved in a separate lawsuit with the Trump administration over an order to stop government agencies from using the company’s AI tools. Former defence secretary Pete Hegseth has publicly labelled Anthropic a “supply chain risk,” a designation that signals government systems should not rely on the product.

For customers, the practical lesson is simpler: access to general-purpose AI can be revoked overnight when the buyer is not the one paying the full cost of a failure.

On Friday, Anthropic’s most advanced models were not updated or fixed in public. They were switched off.