No child touches a tool: why the primary courses are built that way
The safeguarding position behind a catalogue of primary AI courses in which the adult drives and the children think.
The question I am asked most often by primary heads is whether the children will be using AI tools on our courses. The answer is no. No child holds an account, and no child touches an AI tool, on any primary course we run. The adult drives and the children think. The reasons are worth setting out in full, because they are not only about safety, and because this design is the point of the courses rather than a restriction placed on them.
The safeguarding case
There are four parts to it. None of them is dramatic on its own, and together they are decisive.
Age terms. Most general-purpose AI services set a minimum age in their own terms of use. Thirteen is common, and some set it higher. A primary classroom sits entirely below that line, so a school that puts pupils in front of such a tool is relying on an arrangement the provider did not design for them and has, in its own terms, excluded.
Data. Whatever a child types into a tool is data about a child, and it travels to a provider under that provider’s terms. For a primary school that amounts to a decision about children’s data taken at the keyboard, by the child, with nobody else involved in it.
Contact. A chatbot converses. A system that converses with a seven-year-old is a form of contact, and contact is what safeguarding exists to manage. That belongs with the designated safeguarding lead, and not inside a lesson plan.
Guidance. The 2026 edition of Keeping Children Safe in Education, which is the guidance every school in England is required to follow, refers to generative AI and to chatbots, which places these questions within that guidance rather than leaving them to individual preference. Schools do not need to read the guidance heroically. Adult-driven use means there is nothing to authorise, no accounts to audit, no data leaving the school, and nothing for the designated safeguarding lead to sign that they would prefer not to sign.
None of this is a criticism of schools that have chosen differently, and some have done so thoughtfully. The design removes a burden; it does not pass judgement on those who have chosen to carry it.
What a lesson looks like when the adult drives
Take the session How AI Systems Guess. The class begins with a guessing game: predict the next word of a sentence, the next picture in a sequence, the ending of a story, each time from less information than the last. The teacher then runs one demonstration, on a staff account the school has approved, projected to the whole class, using prompts prepared before the lesson, and the class predicts what the system will say before it says it. They are usually right, and when they are wrong they want to know why. By the end of the lesson the class should be able to explain that the system guesses from what it has seen before, and that it has not seen everything.
Or take Fairness and AI: Who Gets Left Out. The class sorts a set of pictures in the way a system might have been taught to sort them, and then looks at what the set was missing: whose faces, whose names, whose languages, whose kind of family. On a whiteboard they work out what a system trained on that set would get wrong, and who would pay for the mistake. No account is opened and no tool is touched, and the children come away with a better working understanding of bias than many adults carry.
The same rule holds across all six primary courses, from What Is AI? for the youngest to Smart and Safe Online with AI for the oldest. Where a tool is used at all, the teacher uses it, on an approved staff account, in front of the class. The adult holds the tool and the children hold the thinking.
What the children leave with
Every primary course ends with something the class has made: a charter, written in their own words, that they can defend. It is three to five statements about when a guessing machine should be trusted, when it should not, and whose work is whose when a machine has helped. A charter might read, in part: a computer guesses, so we check; if it was not shown people like us, it will get us wrong; if the computer helped, we say so. It goes on the wall. Visitors are invited to argue with it, and the class is expected to argue back.
A charter is not a worksheet. It is evidence that the children reasoned, and it is the artefact a headteacher can show to a governor, a parent or an inspector who asks what the school teaches about AI.
Why this is the point rather than a restriction
What a child aged five to eleven needs to learn about these systems is how they guess, who gets left out, and whose work it is when a machine has been involved. None of that requires an account, and all of it requires a teacher. A tool in the child’s hands would get in the way of the thinking, in much the same way that a calculator gets in the way of learning what multiplication is.
The secondary and adult courses are built differently, because the learners are different and the position changes with them. For primary, the design is not a compromise between learning and safety. It is what learning about these systems looks like at that age.
Five questions for any vendor
If you are visiting stands this term, five questions will sort the offers quickly. Do the children need accounts? What data leaves the school, and to whom does it go? What does the child make? Who teaches it, and what have they taught before? What does the designated safeguarding lead have to sign off? The answers to the first question and the last are the ones to listen to most carefully.
Written by the people who do the work at EGM Education and Training. All Insights · Write to the company