Anthropic has temporarily suspended access to its Fable 5 and Mythos 5 AI models after receiving an export control order from the US government.

The company said the directive requires both models to be unavailable to foreign nationals worldwide, including foreign-national employees working at Anthropic. Access to the company’s other AI systems has not been affected.

According to Anthropic, the order was issued on June 12 and cited national security authorities. However, the company said officials did not explain in detail what specific risks led to the decision.

Anthropic believes the concern centers on a reported jailbreak technique that could bypass some of Fable 5’s safety protections. After examining the method, the company said it found only a small number of minor vulnerabilities that were already known and considered manageable.

The company also argued that similar issues can be identified by other publicly available AI models without using jailbreak methods. In its view, the findings do not point to a unique weakness in Fable 5.

Before releasing the model, Anthropic carried out safety testing with government agencies, independent evaluators, the UK AI Safety Institute, and internal teams. The company said the testing showed Fable 5 had stronger safeguards than earlier models and did not uncover any jailbreak capable of broadly defeating its protections.

At the same time, Anthropic acknowledged that no AI model currently offers complete protection against jailbreak attempts. The company said the entire industry continues to face this challenge.

To reduce risks, Anthropic relies on a combination of safeguards, monitoring systems, and policies designed to detect suspicious activity and limit misuse.

The company said it has not seen evidence that the reported jailbreak caused harm. According to Anthropic, the examples shared either produced harmless outputs or highlighted relatively minor software flaws.

Anthropic also said officials only provided verbal information about a possible jailbreak involving code analysis and vulnerability discovery. The company believes the capabilities described are already available in several leading AI systems and are commonly used by security researchers.

While complying with the government order, Anthropic questioned whether a limited jailbreak finding should be enough to remove access to a commercial AI model used by a large number of people.

The company said governments should be able to step in when an AI system poses serious risks, but argued that such actions should be based on a clear process, transparent evidence, and technical facts.

The suspension comes shortly after Anthropic publicly supported stronger oversight of advanced AI systems. In a recent proposal, CEO Dario Amodei called for independent testing of frontier AI models for risks related to cybersecurity, biological threats, loss of control, and advanced research capabilities.

Anthropic said it is working with authorities to restore access to Fable 5 and Mythos 5 and believes the concerns behind the order may stem from a misunderstanding of the reported vulnerabilities.

Also Read: Anthropic Temporarily Bans OpenClaw Creator from Accessing Claude