The US Government Gave Anthropic 90 Minutes to Shut Down Its AI

Emergency export controls enacted

Commerce Secretary Howard Lutnick sent an "informed letter" on a Friday evening requiring individually validated licenses for any foreign national to access the models, citing potential intelligence risks.

Deemed export rules trap engineers

The order applied to foreign employees working inside US offices under "deemed export" rules, forcing Anthropic to lock its own non-citizen engineers out of systems they built.

Global kill switch activated

Unable to instantly verify citizenship status of hundreds of millions of users, Anthropic shut down Fable 5 and Mythos 5 worldwide to avoid criminal penalties.

Amazon CEO reports competitor

Amazon CEO Andy Jassy—whose company has invested $13 billion in Anthropic but also competes with its own AI models—reported a "dangerous jailbreak" to the White House as a "trusted partner."

Debugging treated as cyberweapon

The alleged exploit consisted simply of asking the model to read code and fix software flaws—standard functionality that Anthropic noted is "the product working exactly as advertised."

Cybersecurity industry protests

Over 100 cybersecurity executives from Google, Zoom, and Sophos signed an open letter stating the shutdown confiscated their best defensive tool, noting Fable 5 was already so cautious it refused to read innocuous IT blog posts.

Trillion-dollar IPO jeopardized

The shutdown occurred while Anthropic confidentially filed for an IPO at a $965 billion valuation (up from $380 billion in February), with a $47 billion revenue run rate.

Banks treat US AI as operational risk

JP Morgan Chase and Goldman Sachs immediately cut off Anthropic access for Hong Kong staff, determining that arbitrary Washington control made American AI an operational liability.

G7 summit faces awkward reality

At a French summit days later, President Macron warned that US ability to "turn off the switch" unilaterally damages Western AI leadership, as the "AI3" (Amodei, Altman, Hassabis) met with leaders while their products were frozen.

Treebeard's revenge

Just 24 hours before the letter arrived, Amodei published an essay comparing government AI oversight to "rousing Treebeard" from Lord of the Rings, arguing for statutory power to block unsafe deployments.

Safety measures backfire

Anthropic's "constitutional AI" framework made Fable 5 so aggressively cautious that it often refused basic engineering tasks, yet still triggered emergency export controls.

Perfection demanded, not monitored

Anthropic acknowledged no model is perfectly jailbreak-proof and advocated monitoring over prevention, but the administration treated imperfect safety as justification for immediate shutdown.