All essays
Future of EducationAI in SchoolsLearning VisibilityStudent Wellbeing

The Ten Countries That Agreed

On June 30, UNICEF's Innocenti office released a first-of-its-kind snapshot of what children in ten unlike countries do with artificial intelligence. Twenty million have used it. Thirteen million use it for schoolwork. Two million, one in ten of those surveyed, turn to it for advice about things that worry them. The ten countries did not resemble one another. The finding was the same in every one.

July 4, 20266 min readKoan Team

On June 30, ahead of the first Global Dialogue on AI Governance, UNICEF's Office of Strategy and Evidence in Innocenti released a snapshot brief with a quietly unsettling shape.1 The report drew on survey data from ten countries (Armenia, Brazil, Colombia, the Dominican Republic, Jordan, Mexico, Montenegro, North Macedonia, Pakistan, and Serbia) with roughly a thousand internet-using children ages twelve to seventeen in each, and about a thousand of their parents.2 Ten school systems that agree on almost nothing else. Different languages, different curriculums, different budgets, different histories with the tools of the pandemic and the years after.

The finding was that at least twenty million children in these countries have already used artificial intelligence, adopting it at a pace more than three times faster than the adults around them.3 Thirteen million said they use it for schoolwork. Two million, roughly one in ten of the children UNICEF reached, said they turn to it for advice about things that worry them.1 The last of those numbers is the one worth sitting with. It does not describe a homework habit. It describes a moment in a bedroom, at some late hour, when a child had a small hard thing on her mind and did not, or could not, take it to a person.

What the Ten Countries Actually Had in Common

The ten were chosen for reach across regions rather than for likeness. Armenia's school system is not Brazil's. Pakistan's is not Serbia's. A Ministry of Education framework in one place is a working paper in another. And yet a child in each of those ten countries, when handed a smooth conversational interface at some point in her Tuesday night, produced roughly the same behavior. She used it to help her with her homework. And some meaningful share of the time, she used it for something more tender than that.

The temptation, reading a study like this, is to sort children into risk categories. UNICEF resisted that framing, and so did most of the coverage.3 The report did not conclude that the model was a poor confidant. It concluded that the model was, for a growing minority of children in ten unlike countries, becoming the first confidant. UNICEF's careful phrase for what this needs is stronger safeguards.3 The phrase is right. What the phrase does not say, and what the study leaves unsaid on purpose, is what the child's teacher would need to know if she wanted to notice.

What the Teacher Would See, If Anything

Think about the Tuesday. A twelve-year-old in Belgrade opens a chat window at ten p.m. and asks the model about something worrying her. Perhaps it is about a friendship. Perhaps it is about a paragraph she cannot untangle in her history essay. Perhaps it is both. The exchange goes for eleven minutes. She closes the tab. She turns in the essay on Wednesday. The paragraph is neat. Her teacher grades it and moves on.

Every part of that scene, other than the essay, has left the room. The teacher will not see the eleven minutes. The parent, if the model was used on a school-issued device, will not see the eleven minutes either. In much of the world, no policy exists to keep the eleven minutes. In much of the world, no adult in the child's day will ever learn that she had a question that felt too heavy to ask them, and asked it somewhere else.

UNICEF's brief calls for stronger protections against AI-enabled harms, safer and more transparent systems, improved digital literacy for children and parents, and expanded digital access.3 All of it is right and all of it will take years to arrive in the systems that need it. In the meantime, the question a school could ask itself is smaller, and more immediate. Which parts of a student's week has this school chosen to keep visible?

The Practices That Could Still Be Seen

The revisions to a paragraph, held onto instead of discarded. The pauses long enough to notice which sentences were held longest. The moments where the writing slowed and then quickened again. Those are not surveillance. They are the record of learning that used to sit inside the head of the person across from the child, and that in the era of the neatly finished essay has, quietly, moved somewhere else.

Making that record legible is not a technical problem now. The paper-era habit of throwing away everything except the final page was a habit built for a room where the model did not exist. In a room where two million children in ten countries are already reaching for a model to help hold their worries, that older habit is no longer holding up its side. If the teacher on Wednesday morning cannot see anything of what Tuesday night contained, she is not making decisions about a child. She is grading dust.

What she needs is not the transcript of the chatbot conversation. Most of the children in the study would not welcome that, and neither would their parents. What she needs is a visible record of the child's own practice. The four openings the student tried. The sentence she deleted at eleven-oh-four. The word she paused on for forty seconds. In that record, a careful teacher can read whether the child was working through something or being stuck by it, whether the essay came easily or unusually hard. She cannot rescue every worry. She can see when a child was somewhere unfamiliar without company.

A Future Where the Second Adult Can Still See

Twenty million children in ten unlike countries agreed, without agreeing on anything else, that the interface was easier to ask than the adult. The next decade of school will be a slow negotiation with that fact. The classrooms that manage it will not be the ones with the most restrictive policies. They will be the ones where the second adult, the one the child does actually see the next morning, has been given something to see.

The safeguards UNICEF is asking for are the outer perimeter. The interior work is quieter. It is a room that decides, deliberately, which practices in a student's week it wants to hold onto, and a teacher who has been given time to read them. Neither of those is a policy that arrives in September. Both are choices any school in any of the ten countries could begin making next week.

If a child brought her worry somewhere else on Tuesday night, what would your Wednesday morning look like different if you knew?

References

  1. Children are turning to AI for homework – and life advice

    UN News · June 2026

  2. Snapshot of AI Usage and Concerns Among Children and Parents

    UNICEF Innocenti · June 2026

  3. Children are adopting AI technologies more than three times faster than adults

    UNICEF · June 2026

  4. 20 million children use AI and adopt it faster than adults, UNICEF says

    Euronews · July 2026

Sources cited in order of appearance. Click any inline number to jump.