The US government ordered Claude Fable 5 to be shut down. It cites national security

Anthropic had to deactivate its most powerful language model overnight. This is due to a US government directive citing national security. The company introduced the Claude Fable 5 model as recently as June 10 – so barely two days passed since its launch.

Anthropic had no choice but to shut down the model completely

Anthropic states in its announcement that it received the directive on June 12 at 5:21 PM Washington time (i.e., after eleven PM our time). The order stems from export regulations and mandates suspending access to Fable 5 and Mythos 5 models for all foreign nationals, both outside and inside the US – including the company’s own foreign employees. In practice, this meant that Anthropic had to shut down both models for all customers to comply with the rule. Other models of the company remain available without change, according to its statement.

Dispute over the alleged jailbreak of Claude Fable 5

According to Anthropic, the government justified the move by stating it learned of a method to bypass the model’s security barriers – to “jailbreak” it, so to speak. However, the directive does not contain specific security details, and the evidence, according to the company, was only presented orally. The described technique involves a user presenting specific source code to the model and asking it to fix software bugs. Anthropic claims this is a narrow, not universal, vulnerability, which is also offered by competing models.

The US government ordered Claude Fable 5 to be shut down. It cites national security Adam Kurfürst News Adam Kurfürst News

How Anthropic is defending itself

The company disagrees with the suspension. It argues that withdrawing a commercial model used by hundreds of millions of people due to a narrow vulnerability is disproportionate – and that if the same standard were applied to the entire industry, it would effectively halt the deployment of all new models. Furthermore, according to Anthropic, Fable 5’s protective mechanisms are more effective than any previously deployed model, and no universal way to bypass them has yet been found.

Anthropic stated that it respects the order and is working to restore access. It promised to release further details within 24 hours of its announcement. How regulatory authorities will approach the matter and whether the models will return to users is not yet clear.

Should the government have the authority to order the shutdown of a commercial artificial intelligence model?

Sources: Anthropic, BBC News