Schools Set AI Rules, But Skip Teaching Pupils How to Use It Safely

The numbers behind the gap

Common Sense Media's first annual census on AI use by tweens and teens, published in August 2026 and based on a survey of 1,204 children aged 9 to 17, puts a hard number on something most teachers already sense: pupils are using AI constantly, and schools are struggling to keep pace with anything beyond a warning notice.

Eighty-six percent of children in that age range say they have used generative AI. Among those users, 85 percent apply it to schoolwork, and nearly half do so at least weekly. Sixty-three percent admit they sometimes take an AI-generated answer more or less as given — 25 percent use it as-is, 31 percent rewrite it in their own voice, and 35 percent "improve" it with their own knowledge before handing it in.

The part that should worry school leaders more than the usage figures is what schools have and haven't taught. Seventy-three percent of children say their school told them what they can and cannot use AI for. But only 56 percent say they were taught how to use it safely, and just 51 percent say they were taught how to judge whether an AI answer is accurate or trustworthy at all.

Three in four pupils know the rule. Fewer than half were taught the skill the rule assumes they already have.

That gap matters because a rule without a skill doesn't stop the behaviour — it just drives it underground. Forty percent of teens report using generative AI to help with assignments, and of those, 46 percent say they did it without their teacher's permission. More than four in ten children also say no adult, parent or teacher, has ever talked to them about AI safety at all.

Why a policy statement isn't AI literacy

The instinct in a lot of UK schools over the past two years has been to publish an acceptable-use policy: a paragraph in the behaviour code, a slide in an assembly, a line added to homework instructions. That satisfies the box that KCSIE 2026 and most academy trust governance frameworks ask schools to tick — pupils have been informed of the rules.

But the census suggests that informing pupils of a boundary and equipping them to operate safely within it are different jobs, and most schools are only doing the first one. Knowing you're not meant to paste an essay question into a chatbot is not the same as knowing how to check whether the three facts it gave you back are real, or understanding why an AI tutor's confident wrong answer about a maths method is more dangerous than an obviously nonsense one.

This is also where England's curriculum changes and the DfE's AI tutoring pilots collide with an uncomfortable reality. The government is putting adaptive AI tools in front of pupils — including a pilot aimed at 450,000 disadvantaged Year 9 to 11 pupils from this autumn — on the assumption that pupils will use the tool as designed, under a teacher's framing. The census data says a sizeable minority of pupils are already using general-purpose chatbots their own way, outside any teacher's framing, and most were never taught to evaluate what comes back.

The trust problem this creates

There's a second layer here worth flagging for staffrooms. When 46 percent of assignment-related AI use happens without a teacher knowing, and when three-quarters of schools have set a rule but fewer than half have taught pupils to apply judgement to AI output, teachers are left grading work they can't confidently attribute and pupils are left making safety and academic-integrity calls with tools they were told about but not trained on.

That's not a cheating story so much as a capability story. A pupil who has never been taught to interrogate an AI answer for accuracy will not suddenly develop that instinct the moment a rule is announced. They'll do what feels efficient, which the census shows most already do.

What to do

• Treat AI guidance as a taught skill, not a policy line. A single lesson on "how to check if a chatbot answer is right" will do more than another slide about permitted use.
• Ask pastoral and form-time staff, not just IT or SLT, to raise AI safety directly with pupils — over 40 percent are getting no adult conversation about it at all, at school or at home.
• Build a short, explicit "when and how" into homework instructions rather than a blanket ban, since blanket rules are exactly what the 46 percent unauthorised-use figure grows out of.
• Before rolling out any AI tutoring pilot, check what pupils are told about verifying its output, not just what they're told they're allowed to type into it.

What to watch

Watch whether DfE guidance accompanying the autumn AI tutoring trials includes a pupil-facing critical-evaluation component, or whether it focuses solely on teacher-side controls. If the census pattern holds in England, the next data point worth chasing is how many UK pupils can correctly identify an inaccurate AI answer when given one — a much harder thing to teach than a usage rule, and the one most schools currently skip.