U.S. Orders Anthropic to Block Foreign Access to AI Models

U.S. government blocks foreign access to advanced Anthropic models

Anthropic announced Friday it has received a U.S. government directive requiring the company to immediately suspend access to its Fable 5 and Mythos 5 artificial intelligence models for all foreign nationals. The export control order invokes national security concerns but provides limited specifics about the underlying threat.

According to Anthropic's statement, the company understands the government has identified a method for bypassing safety guardrails in Fable 5—a technique commonly known as "jailbreaking" in AI security circles. The directive requires Anthropic to disable both affected models entirely for all customers to ensure compliance with the order.

Why it matters

This marks one of the first instances of the U.S. government using export control authority to restrict access to specific AI models based on security vulnerabilities. The move signals growing federal concern about advanced AI systems falling into adversarial hands, particularly when those systems can be manipulated to bypass built-in safety measures. For enterprises relying on these models, the incident underscores the regulatory risk inherent in deploying cutting-edge AI capabilities.

Scope and company response

The export control directive affects only Fable 5 and Mythos 5. Anthropic confirmed that access to all other models in its portfolio remains unaffected by the order. The company characterized the situation as potentially stemming from a "misunderstanding" and stated it is working to restore access as quickly as possible.

The abrupt nature of the suspension—requiring immediate compliance—left Anthropic with no option but to disable the models for its entire customer base rather than attempting to screen users by nationality. This approach reflects the practical challenges of implementing granular access controls under emergency government directives.

Broader implications for AI governance

The action against Anthropic comes amid heightened scrutiny of AI model security and the potential for malicious actors to exploit powerful systems. Jailbreaking techniques allow users to circumvent safety filters and content policies, potentially enabling the generation of harmful outputs that developers intended to prevent.

Export controls have traditionally applied to physical technologies and software with clear military applications. Extending this framework to AI models represents an evolving approach to technology governance, one that treats certain algorithmic capabilities as strategic assets requiring protection from foreign access.

The incident also highlights the tension between AI companies' global business models and national security imperatives. As AI systems grow more capable, governments are likely to impose additional restrictions on cross-border deployment and access.

Details of this development were first reported by Reuters on June 12, based on Anthropic's public statement regarding the government directive.