The US government has ordered Anthropic to suspend access to its latest AI models, Claude Fable 5 and Claude Mythos 5. The decision was justified on grounds of national security and concerns regarding the potential for the models’ security measures to be circumvented. As a result, the company has disabled both systems for all users, not just those outside the United States.
According to Anthropic, the US administration has issued an export control directive requiring access to the models to be blocked for all foreign nationals. However, the company claims it has not received detailed information regarding the threat. In its view, the authorities are citing a limited method known as ‘jailbreaking’, i.e. bypassing the model’s security measures, which allows for the identification of a small number of previously known software vulnerabilities. Anthropic argues that other publicly available AI models also possess similar capabilities.
The dispute is significant because it concerns models belonging tothenew‘Mythos’class, which the company had unveiled just a few days earlier. Fable 5 was intended to be a version for general use, equipped with additional security measures. Mythos 5 offered the same capabilities but with limited restrictions in selected areas, including cybersecurity. Anthropic emphasised that these models represent a significant leap in capabilities regarding programming, data analysis and autonomous operation.
Security experts point out that advanced AI models can accelerate the detection of vulnerabilities in IT systems. In the US administration’s assessment, such capability could be exploited for both defence and the conduct of sophisticated cyberattacks. It was this argument that is said to have been decisive in the application of export controls.
The decision sets an important precedent. Until now, US export restrictions have focused mainly on processors and infrastructure used to train AI models. This time, the regulation targets access to advanced software itself. Analysts believe this could mark the beginning of a new phase in global technological competition, in which artificial intelligence will be treated in a similar way to strategic military technologies or semiconductors.
Anthropic, however, disagrees with the US authorities’ assessment. In its statement, the company emphasises that prior to the launch of Fable 5, it conducted several weeks of security testing in collaboration with government agencies and independent experts, and no universal method of circumventing the model’s security has been detected to date. According to the company, the potential vulnerabilities highlighted by the US administration are limited in nature and do not offer capabilities beyond those available in other leading AI models. Anthropic has announced its full compliance with the government’s decision, but considers it disproportionate and based on insufficient technical evidence. The company hopes that this is a misunderstanding and is working to restore access to the models as soon as possible.
