Catalunya's government just said it wants to ban AI in schools for children under 14, part of a wider plan (due as a bill by March 2027) to cut screen time and social media use among kids. The reasoning leans on real research: a Digital Health Guide built from roughly 700 studies, and a fact most Catalan families already live with, teenagers spend close to five hours a day in front of a screen. Poor recent PISA results get partly blamed on screens too.
Writer Saül Gordillo pushed back within days, and his objection is worth sitting with even if you have never set foot in a classroom: "The question may not be at what age we allow students to use AI, but at what age we begin teaching them to understand it." His point isn't that AI in schools is harmless. It's that a flat ban and a real policy are not the same thing, and only one of them actually changes what happens when nobody's looking.
That gap is exactly the one I see business owners fall into. Not with children, with staff.
The ban that isn't a ban
Marta runs a small after-school tutoring academy with three tutors. When ChatGPT started showing up everywhere, she made a rule: no AI on the weekly progress notes parents get, because those notes matter and she didn't want them reading like a form letter. Reasonable instinct. Clear rule.
Except one of her tutors was buried most Thursdays, and a note that used to take fifteen minutes could take two with a quick AI draft. She kept using it. She just stopped mentioning it, because there was nothing to gain by saying so and a rule to break.
Nobody caught this until a parent asked, gently, why her son's note mentioned a topic he hadn't covered that term. The tutor had pasted in a draft without reading it closely enough to notice.
The rule didn't stop the behaviour, it stopped the checking
This isn't a story about one careless tutor. A global 2026 survey of enterprise employees by Lenovo found that roughly a fifth to a third of workers use AI tools their employer's IT department has no visibility into at all. People who have found something that genuinely saves them time on a Thursday do not stop because a policy says so. They stop telling you.
That's the real cost of a blanket "no AI here": not that people obey it, but that the one person positioned to catch a mistake, you, no longer knows to look. The rule removes your oversight before it removes anyone's temptation.
We wrote about the mirror image of this same mistake a few days ago: why making AI mandatory at work can backfire does just as much damage as banning it, for the same underlying reason. A "must use it" order and a "never use it" order both try to settle the question with one word, instead of deciding, task by task, what needs a second pair of eyes before it reaches someone else.
What Marta changed
She didn't lift the ban and hope for the best. She replaced it with one line: AI can draft the first pass of a parent note, but no note leaves the building unless Marta or the tutor who wrote it has read it and can point to one specific detail from that week's actual session.
That single check catches almost everything the ban was trying to prevent: language that could apply to any child, or a fact that simply isn't true this week. It does that without pretending the tool doesn't exist. It also took Marta ten minutes to write down and explain, which is the part a ban skips entirely: a ban feels like a decision, but it's really a way of not making one.
Where an outright "no" is still the right call
None of this means every use of AI deserves a review-and-go-ahead. Some things are worth a flat no regardless of how carefully someone checks the output afterwards: sending anything to a regulator, a bank, or a court under your name is a fair place to keep a person doing the actual drafting, not editing an AI draft, because a subtle error there costs too much even caught late. The real test isn't "AI good, AI bad." It's whether checking the result before it goes anywhere actually removes the risk, or whether the risk was never really about the words on the page.
One thing to do this week
Write down the two or three things at your business you would genuinely never want an AI to touch unsupervised, and name them specifically, not "AI" as one big category. For everything else, replace "don't use it" with "show me before it goes out." You'll catch more mistakes than the ban ever would, because this time you'll actually get to see them.