Anthropic has suspended access to its top-tier AI models, Fable 5 and Mythos 5, for all foreign nationals in response to a directive from the US government.
The order, which was issued last Thursday, applies to both external users and Anthropic staff who are foreign nationals, regardless of their location.
Go deeper with GlobalData
Access deeper industry intelligence
Experience unmatched clarity with a single platform that combines unique data, AI, and human expertise.
The US government cited national security authorities in invoking export control law. The directive covers any foreign national, including those physically present in the US and foreign national employees of Anthropic.
Anthropic stated: “The net effect of this order is that we must abruptly disable Fable 5 and Mythos 5 for all our customers to ensure compliance. Access to all other Anthropic models will not be affected.”
The company said the order was issued without detailed explanation of the government’s national security concern.
According to the company, the government indicated awareness of a method to “jailbreak” Fable 5, effectively circumventing some model safeguards. Anthropic reviewed a demonstration of the technique and determined it exposed a small number of minor, previously known vulnerabilities.
The company stated these issues are similar to what can be identified in other publicly available AI models without using a bypass.
In the weeks before Fable 5 launched, Anthropic worked with US government agencies, the UK’s Artificial Intelligence Safety Institute (AISI), several private security organisations, and internal red teams to test and challenge the model’s security.
Collectively, these teams spent thousands of hours performing adversarial testing, according to Anthropic. Their assessments indicated that Fable 5’s protective measures are more effective than those seen in previous models.
No universal jailbreak, in which a single method bypasses all safeguards, has been identified during testing, said the company.
Anthropic’s security approach for Fable 5 is said to combine multiple safeguards, monitoring, and compulsory 30-day data retention to help detect and address vulnerabilities. The company said this “defence in depth” policy makes jailbreaks either highly specific or costly to achieve.
Anthropic acknowledged that no model provider currently achieves perfect jailbreak resistance, and that targeted attacks uncovering some cyber information remain possible across the industry.
The company also said it has not been notified of any non-universal jailbreak resulting in harmful outcomes. Completed investigations have only found benign or minor issues with no unique impact on Mythos 5.
Anthropic said the government has provided only verbal evidence of a narrow bypass involving software code analysis, which it claims is a capability widely available from other large language models.
Fable 5 was launched last week for developers and enterprise subscribers via the Claude API and multiple Anthropic subscription plans. The Mythos 5 variant is offered through a limited access system, restricted to selected cyber security defenders, infrastructure providers, and researchers in biology.
Anthropic stated it is complying with the government’s legal instructions and disabling foreign national access to both models. The company described the action as unusually broad for a minor vulnerability.
It said: “We disagree that the finding of a narrow potential jailbreak should be cause for recalling a commercial model deployed to hundreds of millions of people. If this standard was applied across the industry, we believe it would essentially halt all new model deployments for all frontier model providers.”
Anthropic said it is working to restore full access if possible.
