Anthropic has disabled access to its Claude Fable 5 and Claude Mythos 5 models worldwide after receiving a U.S. government export-control directive tied to national security concerns, creating one of the most significant AI access restrictions imposed on a frontier model provider to date.
The company said the directive prohibits access to the two models by all foreign nationals, including non-U.S. citizens working at Anthropic itself, regardless of whether they are located inside or outside the United States. Because Anthropic says it cannot reliably distinguish foreign nationals from other users across all access channels in real time, the company chose to disable both models globally rather than attempt selective enforcement.
As a result, Fable 5 and Mythos 5 are currently unavailable to all customers worldwide, including users who would otherwise be unaffected by the restrictions.
The two models represent Anthropic’s newest Mythos-class systems. Released on June 9, Claude Fable 5 was introduced as the company’s first publicly available Mythos model, featuring extensive safeguards in sensitive areas such as cybersecurity, biology, and chemistry. Claude Mythos 5, meanwhile, was offered only to a small group of trusted partners through the company’s Project Glasswing program and includes fewer restrictions than the public version.
Anthropic strongly disagrees with the rationale behind the government order. According to the company, the restrictions appear to stem from concerns surrounding a reported jailbreak technique that allegedly allowed Fable 5 to perform advanced vulnerability discovery and code analysis tasks.
Following an internal review, Anthropic concluded that the capability in question is not unique to either Fable 5 or Mythos 5. The company stated that similar behavior can be observed in several leading frontier models available today, including systems from competing AI developers.
Anthropic described the situation as a misunderstanding and said it is actively working with government officials to clarify the technical facts behind the affected models. The company believes that regulators may have overestimated the uniqueness of the capability being cited.
The incident highlights the growing challenge governments face as AI systems become increasingly capable. Regulators around the world are trying to determine which model abilities should be considered strategically sensitive, particularly in areas such as cybersecurity, scientific research, autonomous decision-making, and advanced software engineering.
The case also raises questions about how export controls can be applied to cloud-based AI services. Unlike physical technologies, AI models are accessed remotely by millions of users around the world, making it difficult to enforce nationality-based restrictions without broader service disruptions.
Anthropic emphasized that the order applies only to Fable 5 and Mythos 5. All other models remain available, including Claude Opus 4.8, which continues operating normally across consumer and enterprise products.
The company says it is working to restore access as quickly as possible and remains optimistic that the issue can be resolved once regulators review the evidence. Until then, the temporary shutdown of Fable 5 and Mythos 5 serves as a reminder that frontier AI development is increasingly shaped not only by technical breakthroughs, but also by evolving government oversight and geopolitical considerations.
