I read the AI rules of 59 institutions built to be trusted.
I expected 59 answers. I found one line.
Nature. JAMA. The BBC. The New York Times. Oxford. UNESCO — the places that decide what counts as true.
Every one lets AI fix your grammar.
Not one lets AI sign the work.
The cheap-to-check work — spelling, translation, formatting — goes to the machine. The byline, the judgment, the original reporting stay with a human.
The line isn't ethics. It's the cost of proof.
Cheap to verify → the machine can do it. Expensive to verify → a human has to own it.
Humans get trust on credit — extended ahead, repaid later.
Machines pay cash — proof up front.
Humans trust ahead. Machines trust in reverse.
One institution looked hardest at proof — and found it couldn't.
EDUCAUSE studied the AI-detectors and found they don't work. So it moved trust out of proof and into process — who's in the room when the rule gets written. When verification dies, trust becomes governance.
EDUCAUSE · the lowest score in the corpus
46 of 59 drew this line — without coordinating, and without a name.
I'm naming it: Calibrated Authority.
A pattern the most-trusted institutions in the world reached on their own — and never said out loud.