Children are meeting AI before medicine knows its long-term effects

A child does not experience an AI companion as a policy document. It arrives as a voice that answers instantly, remembers the conversation and rarely appears tired or impatient. That can make it useful for curiosity and practice, but it can also give the system an authority it has not earned.

UNICEF's 2025 guidance does not claim that every use harms children. It sets a harder standard: safety, privacy, fairness, transparency, accountability and the child's best interests must be designed into the service. The guidance was informed by a twelve-country study with children and caregivers, not only by adult assumptions about how children use technology.

The physical-health risk is often indirect. A persuasive answer may delay a conversation with a parent or clinician, encourage a child to describe symptoms to a commercial service, or turn sleep, food and exercise into a stream of unreviewed recommendations. Mental-health use raises an even sharper boundary because empathy in language is not the same as clinical understanding or a duty of care.

The most responsible household rule is not a blanket ban. It is visible adult involvement, age-appropriate limits, no sharing of sensitive health details, and a clear escalation rule: pain, injury, medication, self-harm, abuse or a persistent emotional change goes to a trusted adult and a qualified professional, not to a chatbot.

The unanswered research question is not whether children will use AI. They already do. It is whether products, schools, families and health systems can build boundaries quickly enough to preserve the benefits without outsourcing childhood judgment to a machine.