The US government has ordered AI company Anthropic to suspend global access to its advanced Fable 5 and Mythos 5 models due to national security concerns related to a potential "jailbreak" that could bypass safety features, prompting Anthropic to disable the models for all users worldwide while disputing the severity and universality of the identified risk, noting that rival models possess similar capabilities and that the government's evidence was primarily verbal and based on a narrow use case involving code analysis. This directive represents a significant escalation in US export controls, shifting focus from hardware to AI models themselves, and comes amidst ongoing tensions between Anthropic and the US government following the company's refusal to allow its models for military surveillance and autonomous weapons, and ahead of Anthropic's confidential IPO filing, with the company warning that such stringent application of jailbreak standards could impede future AI development across the industry.

The US government has ordered AI company Anthropic to suspend global access to its advanced Fable 5 and Mythos 5 models due to national security concerns related to a potential "jailbreak" that could bypass safety features, prompting Anthropic to disable the models for all users worldwide while disputing the severity and universality of the identified risk, noting that rival models possess similar capabilities and that the government's evidence was primarily verbal and based on a narrow use case involving code analysis. This directive represents a significant escalation in US export controls, shifting focus from hardware to AI models themselves, and comes amidst ongoing tensions between Anthropic and the US government following the company's refusal to allow its models for military surveillance and autonomous weapons, and ahead of Anthropic's confidential IPO filing, with the company warning that such stringent application of jailbreak standards could impede future AI development across the industry.

The US government has ordered AI company Anthropic to suspend global access to its advanced Fable 5 and Mythos 5 models due to national security concerns related to a potential "jailbreak" that could bypass safety features, prompting Anthropic to disable the models for all users worldwide while disputing the severity and universality of the identified risk, noting that rival models possess similar capabilities and that the government's evidence was primarily verbal and based on a narrow use case involving code analysis. This directive represents a significant escalation in US export controls, shifting focus from hardware to AI models themselves, and comes amidst ongoing tensions between Anthropic and the US government following the company's refusal to allow its models for military surveillance and autonomous weapons, and ahead of Anthropic's confidential IPO filing, with the company warning that such stringent application of jailbreak standards could impede future AI development across the industry.

In an unprecedented move, the US government on Friday ordered Anthropic—the AI company behind the Claude family of models—to suspend global access to its two most advanced artificial intelligence systems, Fable 5 and Mythos 5, citing national security concerns. Anthropic complied, abruptly taking both models offline for all users worldwide, but pushed back sharply against the directive, calling it a "misunderstanding" and pledging to restore access as soon as possible.

The US Commerce Department issued an export control directive requiring that access to Fable 5 and Mythos 5 be suspended for all foreign nationals—including foreign national Anthropic employees—whether inside or outside the United States.

To ensure compliance, Anthropic said it had no choice but to disable the models entirely for all customers. Access to the company's other models remains unaffected.

The government's stated concern centres on a potential "jailbreak"—a technique used to bypass safety guardrails built into AI models. Anthropic's official statement said the directive arrived at 5.21 pm ET on Friday (early Saturday morning), without specific details of the national security concern.

Based on its own review, the company said the government's evidence amounted to "verbal evidence of a potential narrow, non-universal jailbreak"—one that essentially involves asking the model to read a specific codebase and identify or fix software flaws.

Crucially, Anthropic said it had independently verified that rival models, including OpenAI's GPT-5.5, are capable of the same task. "We have not even received a disclosure of a concerning non-universal potential jailbreak that led to a harmful result," the company said in its official statement.

The action marks the most significant step taken by any government to directly restrict access to a frontier AI model—a sharp escalation from previous US export controls, which had focused on restricting the chips and hardware that power AI systems, rather than the models themselves.

Fable 5 had only launched this week and was described by Anthropic as a new tier of capability it calls "Mythos-class." The company had spent weeks ahead of the launch working with the US government, the UK's AI Safety Institute, and multiple third-party organisations to red-team Fable's safeguards.

The order lands at a fraught moment in Anthropic's relationship with the Trump administration. The two sides have been at odds since earlier this year, when Anthropic refused to allow the US military to use its models for domestic surveillance and fully autonomous weapons systems—a refusal that led the government to place Anthropic on a Pentagon supply chain blacklist, set to take effect later in 2026. That dispute had only recently shown signs of easing.

Separately, Anthropic confidentially filed for a US initial public offering last month. In its statement, Anthropic warned that if the government's jailbreak standard were applied uniformly across the AI industry, "it would essentially halt all new model deployments for all frontier model providers."

TAGS