Two days ago a company shipped the most powerful AI model ever released to the public.
Friday night, the United States government ordered it shut off. Worldwide. Every user, every country, gone by morning.
The trigger was not an attack. A rival showed the government a way around one of the safety locks. That was enough.
The company is Anthropic. The models are Fable 5 and Mythos 5, the most capable systems it has ever built, live for barely 72 hours. At 5:21 on Friday evening, Commerce Secretary Howard Lutnick sent the CEO a letter placing both under export controls, citing national security. The order, on its face, only barred foreign nationals. But Anthropic cannot separate foreign users from everyone else in real time, so to comply it had to pull the plug on all of them. The most advanced AI on earth went dark for the entire planet because of a sentence in a letter.
Here is the trigger, and it is the part that should stop you.
By Anthropic’s account, the government reviewed a single demonstration in which the model was asked to read a codebase and fix its flaws, and it surfaced a small number of previously known, minor vulnerabilities. That is the capability at issue. Finding bugs in code. The same kind of capability that, two weeks ago, a researcher used to catch a four-year-old hole in Zcash before it could be drained. The thing that makes the model a world-class defender is the exact thing that got it called a national security risk.
And Anthropic’s rebuttal lands clean. It says the identical task runs on OpenAI’s GPT-5.5, which sits under no such control at all, and that defenders already use this technique every day. One company’s model is pulled worldwide. A rival’s model, doing the same thing, stays online. National security is the reason given. Competitive accident is the result delivered.
This did not come from nowhere. The same administration spent the spring trying to brand Anthropic a supply chain risk after the company refused to let its models be used for mass domestic surveillance and autonomous weapons. A judge blocked that. The two sides had only just begun to thaw. Anthropic had filed to go public at a $965 billion valuation. On June 2 the President signed an order giving the government early access to frontier models. Then a rival demonstrated a jailbreak, and the most powerful model in the world was switched off three days after launch.
Step back and the pattern is the one that keeps repeating. A zero-knowledge proof hid a four-year flaw. A clean audit hid a redemption gate. A safety wall the company built to look responsible became the precise lever the state used to flip the switch.
The safest lab in AI built a model too powerful to fully release, warned the world it was dangerous, filed to go public at $965 billion, and got it shut down by its own government over a bug-finding trick a competitor can run untouched.
Anthropic calls it a misunderstanding and says it is working to restore access. As of Friday night, the most powerful public AI on earth is a black screen, and the kill switch turned out to belong to the state.