Your pupils are already using it. The lesson is how to think about it.

AI literacy in a school means teaching pupils what these systems are, where they fail, and how to check them, rather than teaching them which buttons to press. It is closer to source criticism than to IT. A pupil who can say why a confident answer might still be wrong has the transferable skill. A pupil who knows a particular interface will not.

Should schools ban AI or teach it?

Bans fail on contact with a phone in a pocket, and they have a specific side effect: they push the use underground, where nobody can teach into it. A pupil who has been told never to use AI and uses it anyway has no way to ask a teacher whether what it told them was true.

The usable position is narrower than a ban and firmer than a free-for-all. Be explicit about which tasks are assessing the process and which are assessing the product. Using a calculator is fine unless the lesson is arithmetic. The same logic transfers cleanly, and pupils understand it immediately because they already accept it elsewhere.

Can teachers detect when pupils use AI?

Not reliably, and this is the part worth being blunt about. Detection tools produce false positives, and they do not distribute those false positives evenly. Pupils writing in a second language and autistic pupils whose written register is unusually formal are flagged more often, which means a detection-led policy quietly penalises the pupils least able to argue back.

Accusing a child of cheating on the output of a system you cannot inspect is not a defensible position for a school. The better instruments are the old ones: know your pupils' writing, assess process as well as product, and use conversation. A pupil who wrote it can discuss it.

What does a good first lesson look like?

Break it in front of them. Ask it something in your subject where you know the answer is subtle and watch it produce something fluent and wrong. The lesson lands harder from a live failure than from any amount of being told to be careful, and it inoculates against the thing that actually causes harm, which is fluency being mistaken for accuracy.

• Ask it a question you know the answer to, then a question you do not. Notice that the two answers sound identical in confidence.
• Ask it the same question three times and compare. Inconsistency is visible in five minutes and explains more than a lecture would.
• Ask it for sources and check one. This is the single most reliable demonstration available.
• Ask it about something local: your town, your school, a local historical detail. Errors appear fastest where the training data was thinnest.
• Then ask where the words went when they typed them. Most pupils have never considered the question, and it is the beginning of the data conversation.

What about the pupils who are not in the room for this?

AI literacy has the same distribution problem as everything else. The pupils with the most support at home will get this explained twice; the pupils with the least will meet these systems alone and unbriefed. Teaching it in school is one of the few interventions that reaches both.

That is part of why Aevum builds the way it does: reasoning a teacher can inspect, so the conversation about how an answer was reached is possible rather than theoretical. An answer a teacher cannot inspect is an answer a teacher cannot teach with. In development.