The U.S. government has ordered Anthropic to suspend access to its Fable 5 and Mythos 5 AI models due to national security export controls. The directive applies globally to all foreign nationals, leading Anthropic to disable the models. The company disputes reported jailbreak concerns as the rationale, stating its safeguards remain effective and that it is working to restore customer access as soon as possible.
The U.S. government directed Anthropic to suspend access to its Fable 5 and Mythos 5 artificial intelligence models under national security export control authorities. The order requires blocking access for all foreign nationals, including employees, whether inside or outside the United States.
Anthropic received the directive and moved quickly to implement the restrictions for all customers. The company stated that access to its other AI models remains unchanged.
Anthropic maintains the decision is linked to reports of a potential method for bypassing safeguards in Fable 5. The company examined a demonstration and discovered it revealed very few new vulnerabilities.
The firm said the testing for Fable 5 was done for thousands of hours by entities including the U.S. Government and the UK Artificial Intelligence Safety Institute. It admits no universal jailbreak has been discovered but agrees absolute impermeability cannot be achieved by any company.
According to Anthropic, Fable 5 employs a defense-in-depth strategy to minimize jailbreaks and includes monitoring for rapid attack response. Customer data is kept for 30 days to enable research and mitigation.
The company states the government’s case involves only verbal evidence concerning a restricted form of jailbreak. Such abilities have been present in many AI applications currently in use by cybersecurity professionals.
While complying with the order, Anthropic believes the action may set a negative precedent. The company is working to restore customer access and will provide more information.
