US Export Controls on AI Models Face Historical Challenges
Original: Encryption, spyware, and now Mythos: History shows why cyber export control doesn’t work
Why This Matters
Sets precedent for AI export control policy affecting entire industry; outcome will establish regulatory framework other labs must follow.
White House ordered Anthropic to restrict export of AI models Fable and Mythos citing national security concerns on June 19, 2026. The move marks first major test of whether US can contain frontier AI technology through export controls, similar to past efforts with encryption and spyware.
The White House issued an export control directive against Anthropic, ordering the company to restrict access to its powerful AI models Fable and Mythos to US entities only, citing unspecified national security concerns. Anthropic complied within approximately 90 minutes, making both models unavailable to all users for over a week. The ban was triggered by two incidents: Anthropic granted South Korean telecom SK Telecom access to Mythos through its partner program, prompting US official concern about alleged China ties which SK Telecom denies; and Amazon CEO Andy Jassy reported researchers found a way around Fable 5's safeguards, though Anthropic disputes characterizing it as a complete jailbreak, describing it instead as a narrow, already-patched issue. Mythos, launched in April 2026, was previously restricted to approximately 150 vetted companies and government organizations due to its capability for cyberattacks. The article notes this represents a first real-world test of US government ability to contain AI technology through export controls, similar to historical attempts with encryption and spyware with mixed results. The US government's track record on such restrictions dates to the 1990s with Pretty Good Privacy (PGP) encryption software, which the government initially tried to prevent distribution of, fearing it would impede intelligence agency surveillance capabilities.