Last term we sat down with a school that had done everything by the book. The moment ChatGPT became a problem, they blocked it on the network. Then they blocked Gemini. Then a list of forty other domains the IT lead kept in a spreadsheet he updated on Sunday nights.
Six months later their integrity cases were up, not down. The head of academics told us, half laughing, that the spreadsheet had become his weekend hobby and he still felt like he was losing.
He was. Here's why.
The ban never reaches the student
A network block stops a request on the school WiFi. It does nothing about the phone in the student's pocket with a data plan, and it does nothing at home, where most homework happens. So the ban doesn't reduce AI use. It moves AI use to the places the school can't see.
That's the part nobody puts on the slide. Before the ban, a teacher could at least glance at a lab screen and see what was going on. After the ban, the same behavior continued behind a hotspot, and the school's actual visibility went to zero. They traded a problem they could see for a problem they couldn't.
The list always loses
The second failure is mechanical. New AI tools appear weekly, and plenty of them are wrappers with friendly names that don't sound like AI at all. A domain list is outdated the moment you save it. The students share working alternatives in group chats faster than any IT department updates a filter, because there are thousands of them and one of him.
One student put it bluntly in a survey the school ran afterwards: the ban was a puzzle, and puzzles are fun.
The honest kids pay the price
Here's the perverse part. The students who respected the ban lost access to a genuinely useful study tool. They couldn't ask for an explanation of a concept at 11pm, couldn't get quizzed before a test, couldn't get feedback on a draft. The students who ignored the ban kept all of that, plus the essay-writing.
So the ban punished exactly the behavior schools want and barely inconvenienced the behavior they fear. Whatever the goal was, that's the opposite of it.
What the school did instead
They stopped trying to block the category and started governing the behavior. The shift sounds subtle, but it changes everything:
- AI tools stay reachable on school devices.
- Every prompt gets checked before it's sent, on the device itself.
- Tutoring, explanations, and feedback go through.
- Essay-writing, exam answers, and homework completion get blocked with a clear message telling the student why.
- Every blocked attempt lands in a dashboard with the context a teacher needs for a real conversation.
Within a few weeks the conversations changed. Instead of "we suspect AI was involved," teachers were saying "on Tuesday at 2:14pm this prompt was blocked, let's talk about it." Students stopped treating the system as a puzzle, because there was no puzzle. The tool worked for them when they used it honestly and stopped them when they didn't.
The takeaway
A ban is a statement. Governance is a system. Statements feel good for a week and then the spreadsheet starts growing. If your school is currently maintaining a domain list by hand, you already know how this ends, and it's not your fault. The approach was broken before you started.
Allow the learning. Block the cheating. You need tooling that can tell the difference, and that's a solvable problem now.