Your washing machine has more safety rules than the AI at your school

A carton of milk gets inspected. So does a washing machine. Your AI assistant doesn't.

That's what Microsoft's Brad Smith said this week in a BBC interview. He's calling for international rules: mandatory incident reporting, independent testing before an AI product reaches the market, and an emergency brake. Mind you: Microsoft sells AI itself. But the question he raises is a fair one. Especially for education.

An imaginary example

A university of applied sciences deploys an AI assistant for attendance. It notices that a student has missed a lot of classes for three weeks and automatically sends a warning about study progress. Efficient: the study adviser doesn't have to do a thing.

But what if that student was ill? Or is a carer? Or a lecturer filled in the attendance list late? The email has gone out and the student is alarmed. Who is responsible now: the lecturer, the study adviser, IT, the vendor?

Nobody did anything wrong. It's just that nobody ever agreed who steers the AI, what it may do on its own and how you stop it.

The market knows this too

This month Google announced AI agents where each agent gets its own identity, limited permissions, a log of everything it does, and a spending cap at which the agent pauses. That's no coincidence.

A survey by software vendor Undo of 300 leaders at large companies running business-critical software found that 81% had an outage caused by AI coding tools within six months. That's a vendor's survey, in a specific sector. But the lesson is broader: building fast is not the same as building in control.

Three questions for tomorrow morning

1. 01
   Who is the owner?

   One name, not “IT”.

2. 02
   What may the AI do on its own, and what does a person decide?

   Especially when it affects students.

3. 03
   Who can stop it, and does everyone know?

Which process at your school already runs partly on AI or automation? Can you say within ten seconds who has the stop button?

If you have to think about that for longer, you don't have an AI policy. You have hope.