Enterprise

Eight days into the Anthropic export ban, enterprise customers brace for a June 20 refund deadline

Commerce Department's Fable 5 and Mythos 5 shutdown has no end date, a 150-signatory open letter on Lutnick's desk, and a Bloomberg legal read that Washington just claimed kill-switch power over any frontier model.

By Asher Glick · Enterprise software desk · 2026-06-19 · 3 min read

Photo: Unsplash / Andy Feliciotti — The United States Capitol dome in Washington, D.C.

Eight days after a 5:21pm ET directive from Commerce Secretary Howard Lutnick forced Anthropic to pull Fable 5 and Mythos 5 offline for every customer worldwide, the models are still dark, refunds are queued for a June 20 deadline, and Bloomberg’s legal desk is now arguing that Washington has effectively claimed kill-switch authority over any American-made frontier model.

The order, as Anthropic described it, instructed the company to suspend access to both models for any foreign national inside or outside the United States, including its own foreign-national employees. Rather than build a citizenship gate inside its API, Anthropic shut the models off for everyone. The rest of its lineup is untouched.

The legal frame Bloomberg laid out on Thursday is the part enterprise buyers are reading twice. Lutnick stretched technology-transfer law from its traditional perch over chips and weapons systems to cover mere usage of a deployed model. There was no court order. There doesn’t appear to have been one needed. TechCrunch’s read, published four days earlier, was blunter: the government just demonstrated it can force a tech company to take a product offline through swift, unilateral action, and it set that precedent on software rather than hardware.

What triggered it’s, by Anthropic’s account, thin. The Wall Street Journal reported that Amazon CEO Andy Jassy told U.S. officials, including Treasury Secretary Scott Bessent, that Amazon researchers had prompted Fable 5 into producing material that could aid cyberattacks. Anthropic says the government’s evidence, delivered verbally, describes a narrow non-universal jailbreak that amounts to asking the model to read a codebase and patch flaws, a capability it notes is widely available in OpenAI’s GPT-5.5.

The timing is the part nobody at Anthropic can love. Dario Amodei published an essay days earlier calling for binding regulation that included the power to block unsafe models. Both Anthropic and OpenAI have confidentially filed IPO prospectuses.

Alex Stamos, chief product officer at Corridor and the former Facebook security chief, organized an open letter to Lutnick and National Cyber Director Sean Cairncross now carrying more than 150 signatures from executives and technical leaders, asking for the directive to be lifted. Bloomberg’s opinion side is pushing the opposite move: take the Nvidia route, cooperate quietly, don’t litigate in public.

That’s the choice on Amodei’s desk. The essay argued for a regulator with this kind of power. The regulator now has it.

Sources