The story behind Anthropic's shutdown keeps getting interesting.
Last night, the government ordered Anthropic to disable Fable 5 and Mythos 5 worldwide.
The reported trigger: Amazon, one of Anthropic's own biggest investors, told the administration its researchers had used the model to pull cyberattack relevant info.
Except the security researcher Anthropic showed the report to says it wasn't even a jailbreak.
She called it defensive research, the kind defenders are supposed to do, and said the government "scored an own goalโ against national security by pulling the model.
And Anthropic notes the same technique works on other public models, including GPT-5.5, that didn't get touched.
Strip the politics out and one builder lesson is left standing: a hosted frontier model can vanish overnight because of a competitor's report, an investor's phone call, and a Friday night decision.
Reasons you'll never fully see.
The model on your own hardware doesn't care who called whom.
That's an argument for local and open weights, made in a single news cycle.