The directive followed reports that a third-party company had successfully jailbroken Mythos 5 to bypass safety safeguards and access powerful cybersecurity and bug-hunting capabilities. The Pentagon, which maintains a complicated relationship with Anthropic—having previously sought to ban most of its products while carving out an exemption for Mythos for its own use—flagged urgent national security concerns about potential cyberattack misuse. The shutdown occurred just 72 hours after Anthropic's June 9 public launch of Fable 5, marketed as the world's most powerful AI model. Anthropic executives are scheduled to fly to Washington on Monday, June 16, to attempt resolution, though the models remain disabled.
Attorneys should monitor this as a watershed moment: the first instance of the U.S. government effectively shutting down a major public AI product on national security grounds, banning it domestically and globally. The incident exposes fundamental tensions between technological advancement and government security protocols while complicating Pentagon reliance on AI capabilities. A critical open question remains whether the jailbreak vulnerability is unique to Mythos 5 or already exists in competitor models like OpenAI's GPT-5.5, which remain fully operational.