U.S. Imposes Export Controls on Anthropic's Fable and Mythos AI Models
The U.S. government has implemented export controls on Anthropic's advanced AI models, Fable and Mythos, following Amazon researchers' discovery of a method to bypass some of Fable's cybersecurity safeguards. This decision has compelled Anthropic to disable the models for all users, citing American "deemed export" rules. The move is prompting debate within the AI community regarding the future of frontier AI development and has led to concerns in Europe over AI sovereignty, while reportedly pleasing open-source AI developers in China.
The U.S. government has imposed export controls on Anthropic's powerful AI models, Fable and Mythos. This action was taken after researchers at Amazon reportedly found a way to circumvent some of Fable's cybersecurity guardrails, raising concerns about potential vulnerabilities.
As a direct consequence, Anthropic has been forced to disable both Fable and Mythos for all users. American "deemed export" regulations dictate that access by any foreign national, including Anthropic's own employees, would violate the law under these controls. Anthropic has initiated efforts to have the export controls rescinded, sending executives to Washington D.C. for discussions, though no agreement has yet been reached.
The government's decision has generated diverse reactions. Some observers, critical of Anthropic's communication style regarding its models' dangers, reacted with a sense of vindication. Conversely, many cybersecurity experts have argued that the Fable jailbreak discovered by Amazon did not unlock offensive cyber capabilities beyond what is already available in other AI models, such as OpenAI's GPT-5.5, which are not currently subject to similar controls. Over 100 cyber and tech policy experts signed an open letter, asserting that Fable and Mythos are vital tools for cyber defenders to identify and patch system vulnerabilities, and that these benefits outweigh the risks of potential misuse.
Amazon's role in the decision is notable, with CEO Andy Jassy reportedly contacting the White House regarding the issue. Amazon has invested $13 billion in Anthropic and has committed to potentially investing up to $20 billion more in the future. The specific internal deliberations at Amazon, including how the company weighed its significant investment against the national security concerns it raised, remain publicly unclear.
This government action has been interpreted by some as establishing a de facto licensing regime for frontier AI in the U.S. This regime is characterized as ad hoc and opaque, lacking the transparency of a formal, de jure system. Jonathan Iwry, a fellow at the Wharton Accountable AI Lab, stated that the government appears to be repurposing existing legal frameworks for this purpose.
(Source: Fortune)
