Three stories today, from three different kinds of institution, that only look unrelated. A city government, a national court and a medical regulator each drew a line around artificial intelligence this week — and no two of them drew it in the same place.

The line that pulls AI out

New York City, the largest school district in the United States, banned student-facing generative AI for roughly 600,000 children from pre-kindergarten through eighth grade. Mayor Zohran Mamdani's argument was blunt: he has not seen a study, outside the ones funded by the companies selling the tools, showing AI helps young children learn. The instinct here is protective — keep the technology away from developing minds until someone proves it belongs there.

Britain's schools, worth remembering, have gone the other way: the Department for Education treats AI as something to adopt carefully, not lock out. Same evidence, opposite reflex.

The line that clears AI's path

In Manhattan, the opposite instinct won. The US Justice Department filed to back OpenAI in the consolidated copyright cases brought by the New York Times and a dozen other publishers and creators, arguing that training models on copyrighted text is "extraordinarily transformative" and should count as fair use. Where New York protected children from AI, Washington moved to protect the AI industry — on grounds of competitiveness, scientific progress and, tellingly, the argument that strong copyright would only entrench the biggest labs. Britain, mid-fight over the same question and with no preferred answer of its own, has drawn no such line in open court.

The line that lets AI decide

And in Europe, a regulator went furthest of all. Vara's autonomous breast-screening tool became the first of its kind cleared anywhere to pass normal mammograms through without a radiologist reading them. Not flag, not assist — decide that a scan is normal enough that no human needs to look. For an NHS short on radiologists, the appeal is obvious; the risk sits entirely in the scans the machine clears alone.

The through-line

Put together, the week's real signal is not that institutions are restricting AI, or that they are enabling it. It is that they have all decided it is now consequential enough to need a line drawn — and they profoundly disagree on where.

A classroom draws the line at "not yet, not for children." A court draws it at "yes, because the alternative costs us the race." A health regulator draws it at "yes, even for a life-or-death call, as long as we can watch it." Each is defensible. None of them is the others. And in every case the UK is standing slightly off to one side, having not quite decided which room it wants to be in.

That is the actual state of AI governance in September 2026: not a single argument moving in one direction, but a dozen separate arguments, in a dozen separate institutions, all being had at once. The technology arrived faster than the consensus, and the lines are being drawn one room at a time.