US Bars Foreign Nationals From Anthropic's Top AI Models

The US government has ordered Anthropic to cut off access to two of its most advanced AI models for all foreign nationals, including those working inside the United States. The directive, delivered late Friday, forced the company to suspend usage of Fable 5 and Mythos 5 across its entire customer base to ensure compliance.

Anthropic described the move as an export control measure tied to unspecified national security concerns. The order arrived at 5:21 p.m. and offered no detailed explanation of the risks involved. In response, the company disabled the models for everyone rather than attempt to segregate users by nationality on short notice. Other Anthropic products, including its Claude chatbot, remain available without restriction.

Fable 5 was released publicly only four days earlier. It was designed to bring some of the capabilities previously limited to Mythos, a specialized cybersecurity model, into wider use. Mythos itself is reserved for a small set of government and industry partners through Project Glasswing. The newer model demonstrated stronger performance on complex tasks, including game environments that earlier systems had failed to navigate. Both models carry advanced abilities in identifying software vulnerabilities and handling biotechnology-related queries, areas that regulators view as dual-use technologies.

The breadth of the restriction stands out. It applies to foreign nationals regardless of their location or employment status at Anthropic. That includes researchers and engineers already inside the company who hold non-US passports. Such a sweeping approach reflects an expanding view within the government that certain AI systems cannot be safely shared even under controlled corporate conditions.

Anthropic noted it had invested significant effort in safety measures before launching Fable 5. The sudden order suggests officials concluded those steps were insufficient once information about potential jailbreaks circulated. The company itself has linked the timing to reports of methods that could bypass its safeguards on the new model.

This episode illustrates how export controls, once applied mainly to hardware and traditional software, are now being used to manage frontier AI systems. The policy treats advanced models as strategic assets whose spread must be limited even when the underlying company is American. For firms like Anthropic, the result is operational friction: technical talent from abroad becomes harder to integrate fully, and product roadmaps face abrupt external constraints.

The directive also underscores the difficulty of drawing clean lines around AI capabilities. Mythos was already positioned as a defensive tool for finding long-hidden vulnerabilities. Extending similar power to a broader audience through Fable 5 apparently crossed a threshold that prompted immediate government intervention. How other developers will respond, and whether similar orders will follow for competing models, remains an open question for the industry.