The U.S. government ordered Anthropic to block foreign nationals from accessing its Fable 5 and Mythos 5 AI models on Friday, June 14, 2026 [1].

This move signals a tightening of federal control over the export of high-capability artificial intelligence. By restricting access to these specific models, the U.S. aims to prevent foreign adversaries from reverse-engineering or exploiting advanced AI capabilities that could be used for cyberattacks or intelligence gathering.

Anthropic, a research company based in the United States, complied with the directive by suspending access for users outside the country [1]. The restriction affects two specific models [2], Fable 5 and Mythos 5.

"Anthropic has blocked access to its Fable 5 and Mythos 5 models for foreign users following a directive from the U.S. government," an Anthropic spokesperson said [1].

The federal government cited specific technical vulnerabilities as the driver for the order. Officials expressed concern that foreign actors could use "jailbreak" attempts to bypass the safety guardrails of the models, potentially exposing sensitive capabilities that could jeopardize national security [3].

"The decision was made out of national-security concerns, particularly the risk of jailbreak attempts against our models," a U.S. Department of Commerce official said [3].

Anthropic has not detailed how it will verify the nationality of users or which specific countries are targeted by the block. However, the company indicated it will continue to follow federal mandates to ensure its technology does not fall into unauthorized hands.

"We remain committed to complying with all applicable laws and regulations," the Anthropic CEO said [3].

“The U.S. government ordered Anthropic to block foreign nationals from accessing its Fable 5 and Mythos 5 AI models”

This directive reflects an evolving U.S. strategy to treat frontier AI models as dual-use technologies, similar to advanced semiconductors or military hardware. By targeting 'jailbreak' risks, the government is acknowledging that software-level safety filters are insufficient to protect national security interests against determined foreign state actors, necessitating hard geographic and identity-based access controls.