Anthropic has announced its plans to disable access to its latest top-tier AI models, Fable 5 and Mythos 5, in compliance with a US government export control order that restricts foreign individuals and entities from using these products.
In a blog post published on Friday evening, the company revealed it received an official letter from the US government at approximately 5:21 p.m. ET, citing national security concerns regarding #Anthropic's models. The scope of this mandate covers any foreign national inside or outside the US, including "foreign national Anthropic employees." Consequently, the net effect is a complete disabling of the models for all users to guarantee regulatory compliance.
While the government's communication lacked specific details about the national security risks, Anthropic believes the concern centers around a potential "jailbreak" vulnerability in Fable 5. However, the company disputed the severity of the issue, stating that the technique is narrow rather than universal, and involves known vulnerabilities that are easily detectable by other publicly available AI models.
This development represents a major escalation in the ongoing conflict between Anthropic and the Trump administration over AI safety, national security, and government oversight of frontier models. Previously, in February, the Pentagon designated Anthropic as a "supply chain risk" after the startup sought restrictions on deploying its models for certain military applications.
In retaliation, Anthropic filed lawsuits against the Department of Defense. Currently, two lawsuits contesting the supply-chain risk label remain pending. Although Anthropic disagrees with the government's findings, it confirmed it is complying with the directive. A spokesperson declined to specify the exact date when access to Fable 5 and Mythos 5 will be cut off, but noted that other Anthropic models will remain accessible.
[AgentUpdate Depth Analysis] This forced shutdown of Anthropic's flagship models highlights the growing geopolitical vulnerability of the global AI Agent ecosystem. Modern AI Agents rely heavily on borderless API access to leading frontier models to execute complex, multi-agent workflows. By restricting access based on nationality, these regulatory crackdowns fragment global collaboration and threaten the reliability of agentic automation. Consequently, developers are likely to pivot away from heavily regulated US proprietary APIs, accelerating the adoption of open-source architectures like Llama and decentralized runtime environments. For the Agent ecosystem, regulatory compliance and local model hosting will transition from secondary considerations to core architectural requirements.
