“Today's children are the first generation to grow up with artificial intelligence – yet often the last to be asked for their views.” That was Volker Türk, the UN High Commissioner for Human Rights, speaking in New York on 22 September at a high-level meeting on protecting children in the age of AI.
His office and partners had surveyed more than 1,000 children across 49 countries. Their worries were specific: fake images made of real people, personal data harvested and exploited, creativity quietly eroded, and AI edging out real relationships with family, friends, teachers or therapists.
A day later, at a side event co-hosted by UNICEF, a second message came through. Lily Caprani, UNICEF's Chief of Global Advocacy, said evaluating AI for children has to go further than preventing the worst harms. The aim, she said, is to make sure children using AI in their daily lives are “actually learning and flourishing”.
Neither event created rules for schools. Both describe habits a school can adopt now.
The children's list is a curriculum
Look again at what the children named. Fake images of real people is a lesson about synthetic media, consent and how to respond when it happens to someone you know. Data harvesting is a lesson about what a free tool collects and why. Eroded creativity is a question about when to draft without help. Replaced relationships is a conversation about what a chatbot can and cannot be.
The stakes are not abstract. UNICEF research cited at the meeting found that at least 1.2 million children in 11 countries had their likenesses turned into AI-generated sexual images in the past year. For schools, that sits in safeguarding as much as in the curriculum, and pupils need to know who to tell and what will happen next.
Children are not waiting for adults to finish writing the rules. They are already using these tools. Starting AI literacy from their concerns, rather than from a list of products, keeps the lessons relevant and gives pupils a reason to engage.
Judge tools on learning, not just safety
UNICEF's Chief AI, Henrietta Ridley, set out three points about evaluating AI for children. Children do not experience AI in the same way, so evaluation has to account for age, language, culture and circumstance. Some risks only appear over time, so a single test conversation is not enough. And safety alone is not enough: the question is also whether a tool supports children's development and wellbeing.
That gives a school a better set of questions for any vendor than “Is it safe?” Ask what evidence there is that pupils learn more with the tool than without it. Ask how it behaves with a younger pupil, a pupil learning in a second language, a pupil who uses it every day for a term. Ask who outside the company tested it.
Ask pupils before you roll out
The coalition behind the New York meeting wants children involved in how AI systems are designed, deployed and monitored. A school cannot redesign a product, but it can involve pupils in decisions about how AI is used in their own classrooms.
Before introducing a new tool, run a short conversation with a class or the school council. What do they already use? What worries them? Where do they think AI would help them learn, and where would it do the work for them? Their answers will sharpen your policy, and the conversation itself is AI literacy in practice.
Türk told governments they must do more than encourage good behaviour: “Voluntary commitments are not enough.” Turning that into regulation will take time. Schools do not need to wait for it. Asking pupils first costs a lesson.
Sources: UN News, ‘Ask us first’: Children demand a voice on AI's future, 22 September 2026; UNICEF, Putting children at the centre of independent AI evaluation, 8 October 2026. The coalition's statements are commitments of intent, not binding rules.