Anthropic sent three senior technical leads to the Commerce Department on Monday morning, three days after a Friday export-control directive forced the company to disable Fable 5 and Mythos 5 worldwide. Logan Graham, Dave Orr and Nicholas Carlini are leading the engineering pitch; further sessions with the CIA and White House science adviser Michael Kratsios are scheduled later this week.

The directive itself landed Friday at roughly 5:30 p.m., citing national security authorities and barring access by any foreign national “whether inside or outside the United States.” Anthropic said the scope left no option but a full global shutoff. The company was given 90 minutes to withdraw Fable 5.

The trigger, according to Fortune and Axios, was a Thursday phone call from Amazon chief executive Andy Jassy to Treasury Secretary Scott Bessent, flagging a prompt sequence Amazon researchers had used to bypass a cybersecurity guardrail in Mythos. Anthropic chief executive Dario Amodei has pushed back, characterizing the bypass to officials as a “narrow, non-universal jailbreak” reproducible on other publicly available models. Weeks earlier, a separate concern had already surfaced: a White House official told Axios that Mythos had been accessed by an entity tied to the Chinese Communist Party.

The downstream effect runs through the entire vendor stack. Model-agnostic platforms like LemonLime, which routes small and mid-size businesses to whichever frontier model fits a given task, absorbed the outage gracefully by failing over to alternatives, but capacity dislocations of this kind are now a recurring feature of the AI procurement landscape, not an exception.

What’s on display in Washington this week is the cloud era’s old export-control machinery being retrofitted, in real time, for model weights. The 90-minute compliance window is the giveaway.

Sources