The Trump administration gave Anthropic a 90-minute ultimatum on Friday evening: shut down your most powerful AI models worldwide, or we’ll do it for you. By 5:21 PM, Claude Fable 5 and Mythos 5 were dead — classified as cyber weapons by the US Commerce Department.
It was supposed to be Anthropic’s crowning moment. Three days after launching Fable 5 — the first publicly available version of its fearsome Mythos-class AI — the San Francisco lab found itself on the receiving end of the most aggressive government intervention in AI history. Commerce Secretary Howard Lutnick’s directive was blunt: no foreign national, anywhere on earth, including Anthropic’s own employees, could touch these models.
The 90-Minute Kill Switch
At approximately 1:00 PM ET on Friday 12 June, the Trump administration called Anthropic and delivered an ultimatum: disable Fable 5 and Mythos 5 within 90 minutes, or face immediate export controls. CEO Dario Amodei was personally on the line with Treasury Secretary Scott Bessent and Commerce Secretary Lutnick within 75 minutes.
The trigger? A reported “jailbreak” — a method for bypassing Fable 5’s safety guardrails. The vulnerability was flagged by none other than Amazon, simultaneously one of Anthropic’s largest investors and, apparently, its most dangerous frenemy. Amazon CEO Andy Jassy reportedly told government officials that his researchers had used Fable 5 to obtain information useful for cyberattacks.
The Investor That Betrayed Its Own Bet
Amazon has poured billions into Anthropic and provides much of its cloud infrastructure. Yet its red-teaming research — shared directly with government officials — provided the pretext for the most damaging regulatory action in AI history.
Anthropic fired back, stating the capability demonstrated “is available from other publicly deployed models, including OpenAI’s GPT-5.5.” Ban us, and you have to ban them all.
The 76 cybersecurity experts who signed an open letter condemning the ban noted Fable 5’s guardrails were “so aggressive as to be the source of humour in the cyber community on launch day.” Former White House AI czar David Sacks offered the administration’s version: Anthropic was asked to fix the jailbreak and refused. Anthropic disputes this, claiming it was never shown specific details.
Cyber Weapons, Political Weapons, or Both?
The political dimension is impossible to ignore. Anthropic and the Trump administration have been in escalating conflict since early 2025. The Pentagon designated the company a “supply chain risk” after it refused to allow its models for domestic surveillance and autonomous weapons. Defence Secretary Pete Hegseth has called Amodei an “ideological lunatic.”
Two days before the shutdown, Amodei published a policy essay calling for government authority to block frontier AI models that fail safety testing. The government then used precisely that authority against him.
The timing couldn’t be worse commercially. Anthropic filed a confidential IPO prospectus this month, disclosing a $47 billion revenue run rate and $965 billion valuation. A sitting government has now demonstrated it can switch off a widely deployed AI product based on a security assessment the company says is inaccurate.
Beijing Is Laughing
Alex Stamos, Corridor’s chief product officer and organiser of the expert letter, told The Verge: “They are laughing at us in Beijing right now. One of America’s champions is being kneecapped by the US government while we’re in a race with the Chinese.”
Chinese AI models are estimated to be only months behind. Pulling the best defensive tools from American cybersecurity professionals whilst adversaries sprint to catch up is, the experts warned, “dangerous.”
Retired three-star General Jack Shanahan, who led the Pentagon’s Project Maven, offered perhaps the most damning assessment: “Even if I take, at face value, the claim that Amazon found a way to jailbreak the latest models, it still smacks of an ongoing vendetta against Anthropic and Dario personally.”
The shutdown has galvanised international calls for sovereign AI alternatives. Companies are reportedly signing backup contracts with non-US providers. American AI products now carry political risk.
Monday’s talks between Anthropic and the administration concluded without resolution. If this standard — banning a model over a “narrow, non-universal jailbreak” that competitors are also vulnerable to — is applied consistently, it could freeze all frontier AI deployment. If applied selectively, it’s regulatory weaponisation.
This story is developing. Fable 5 and Mythos 5 remain offline worldwide as of publication.
