Anthropic Blocks New AI Models for Non-U.S. Users Over National Security Concerns

—

in

The U.S.-based artificial intelligence company Anthropic announced on Friday, June 12, 2026, that it has disabled access to its newly launched AI models — Fable 5 and Mythos 5 — for all non-U.S. users. The move follows an export control directive issued by the U.S. government, specifically the Bureau of Industry and Security (BIS), which enforces strict restrictions on the export of frontier AI models and advanced chips — such as those made by Nvidia — to certain countries, including China, Iran, and Russia, citing national security concerns.

Government Directive Affects Global Access

In a statement, Anthropic confirmed that the U.S. government, invoking national security grounds, ordered a halt to *all* access to Fable 5 and Mythos 5 by foreign nationals — whether located inside or outside the United States, including Anthropic’s own non-U.S. employees. The company emphasized that access to all its other AI models remains unaffected.

Calling the situation the result of a ‘misunderstanding,’ Anthropic assured users it is working urgently to restore access to the two models as soon as possible.

Jailbreak Concerns Spark Regulatory Action

According to Anthropic, U.S. authorities believe a method exists to ‘jailbreak’ Fable 5 — that is, bypass its built-in safety safeguards. However, the company stated that the reported issues were already known, minor in nature, and replicable using other publicly available AI models.

Prior to launching Fable 5, Anthropic said it conducted thousands of hours of joint safety testing with U.S. government agencies, the UK’s AI Safety Institute, and other independent groups. It added that no party had identified a way to fully circumvent the model’s safety systems.

The company noted that U.S. officials provided only oral details about a single, narrow workaround: asking the model to review a specific codebase and fix software vulnerabilities. Anthropic pointed out that this capability is widely available across other leading models — including OpenAI’s GPT-5.5 — and is routinely used by cybersecurity professionals to maintain secure systems.

Broader Implications for the AI Industry

Anthropic warned that expecting perfect jailbreak resistance from any AI model is currently unrealistic. Applying such a standard across the industry, the company argued, would effectively halt all development of new frontier models by every major provider.