Anthropic suspends Fable 5, Mythos 5 access after US export control directive

Anthropic defended the safety measures implemented in Fable 5, noting that the model underwent extensive testing by US government agencies, the UK's AI Safety Institute, third-party organisations and internal teams before launch.Anthropic has been forced to suspend access to its Fable 5 and Mythos 5 artificial intelligence models after receiving an export control directive from the US government citing national security concerns.In a statement issued on Friday, following the order, the company said the order requires it to disable access to the two models for all foreign nationals, both inside and outside the United States, including Anthropic employees who are not US citizens.

As a result, the company said it must suspend access for all customers to ensure compliance with the directive.Access to Anthropic's other AI models will remain unaffected.According to the company, the directive was received at 5:21 pm Eastern Time and did not include specific details regarding the underlying national security concerns.Anthropic said its understanding is that authorities became aware of a method of bypassing, or "jailbreaking", Fable 5's safeguards.

The company reviewed a demonstration of the technique and concluded that it exposed only a small number of previously known vulnerabilities."The vulnerabilities all appear relatively simple, and we have found that other publicly available models are able to discover them as well without requiring a bypass," the company said.Anthropic defended the safety measures implemented in Fable 5, noting that the model underwent extensive testing by US government agencies, the UK's AI Safety Institute, third-party organisations and internal teams before launch.The company said those evaluations demonstrated that Fable's safeguards were substantially stronger than those of previous generations of AI models and that no testers had identified a universal jailbreak capable of broadly bypassing the system's protections.Dispute over 'jailbreak' findingsAnthropic acknowledged that perfect resistance to jailbreak attempts may not be achievable for any AI model currently available and said it adopted a "defence in depth" approach that combines safeguards, monitoring, and mitigation mechanisms.The company also pointed to its policy of retaining customer data for 30 days on Fable 5, arguing that the measure enables rapid investigation and response to potential security issues.Anthropic further claimed that the vulnerabilities disclosed to it were either benign or involved capabilities already widely available in competing AI systems.While complying with the order, Anthropic criticised the government's decision, arguing that the discovery of a narrow, non-universal jailbreak should not be sufficient grounds to withdraw a commercial AI model used by hundreds of millions of people.The company warned that applying such a standard across the AI industry could significantly slow or halt the deployment of future frontier models.Anthropic said it supports government oversight of advanced AI systems but argued that intervention should be based on transparent processes, clear technical evidence and consistent standards.