The US government has ordered Anthropic to suspend access to its most advanced AI models, Fable 5 and Mythos 5, for all foreign nationals, citing national security concerns. The company received an export control directive without being given specific details about the security issues involved.
Anthropic announced it will "abruptly disable" these models for all users in response to the government order. The restrictions mark a significant shift in how the US regulates access to cutting-edge artificial intelligence technology, applying export controls typically reserved for physical hardware to AI systems themselves.
The move comes after Anthropic disclosed that unauthorised users gained access to its Mythos model through a third-party vendor environment. Bloomberg reported that a small group of people in a private online forum accessed the model on the same day Anthropic began releasing it to select companies including Apple and Goldman Sachs for testing. One of the unauthorised users worked as a contractor for Anthropic and deployed methods used by cybersecurity researchers to gain access. According to Bloomberg, the group was primarily interested in "playing around" with the technology rather than causing harm, and had not run cybersecurity tests on the model.
The incident raised alarms among government authorities who have expressed serious concerns about Mythos's capabilities. The UK's AI Security Institute warned that Mythos represents a "step up" from previous models in terms of cyber-threat potential. The model can identify software vulnerabilities and conduct attacks requiring multiple actions without human intervention, tasks that would normally take human professionals days to complete. Mythos successfully completed a 32-step cyber-attack simulation created by the security institute in three of ten attempts.
Kanishka Narayan, the UK's AI minister, cautioned that UK businesses "should be worried" about the model's ability to detect flaws in IT systems that hackers could exploit.
Anthropic's CEO stated that research leading to the US export restrictions was conducted for defense and cybersecurity purposes. The CEO emphasised that the work was not a jailbreak of the AI systems but rather legitimate security testing.
The government's decision to impose export controls on AI model access rather than just underlying hardware represents a new approach to technology regulation. As AI companies continue developing increasingly capable systems, US officials are taking a more proactive stance on controlling access to advanced tools. The restrictions aim to maintain American advantages in AI development while preventing potential adversaries from accessing systems that could be used for military or intelligence purposes.
The move could set a precedent for how other leading AI systems are regulated in the future as national security concerns around advanced AI capabilities intensify.