Child-Safe AI: A Framework for Learning and Privacy

AI built for children should do three things at once: help them think for themselves, protect their personal data, and be honest that it is a machine. Regulation sets a floor for privacy, but it says little about learning, so the people who design, buy and use these tools need a fuller framework.

This article proposes six principles, drawn from UNICEF’s guidance, children’s privacy law and recent research on AI and learning. It is written for developers, schools and parents choosing AI for children.

What should child-safe AI achieve?

UNICEF’s Guidance on AI and children, updated to version 3.0 in December 2025, sets ten requirements for child-centred AI. They include ensuring safety, protecting children’s data and privacy, providing transparency and accountability, supporting children’s best interests, development and well-being, and preparing children for a world with AI. The update adds attention to AI companions used by children.

Those requirements are broad. The six principles below turn them into questions a team or a parent can check.

1. Does it build thinking rather than replace it?

An AI that answers every question instantly can stop a child from doing the work that makes learning stick. In a field experiment with nearly 1,000 high school students, Bastani and colleagues found that an unrestricted chatbot improved practice but harmed later exam results, while a tutor that gave hints rather than answers avoided the harm.

In practice: the AI should ask questions back, give the next hint rather than the solution, and let adults set how much help it gives.

2. Does it collect as little data as possible?

Children’s data deserves stricter care than adults'. The US COPPA Rule requires verifiable parental consent before collecting personal information from children under 13, and says that information may be kept only as long as is reasonably necessary for its purpose, never indefinitely, under a written retention policy.

In practice: collect only what the learning needs, publish how long it is kept, delete it automatically, and never use it for advertising or to build commercial profiles.

3. Is it honest that it is a machine?

Children can form attachments to AI that talks like a friend. In 2025 Common Sense Media concluded that social AI companions pose unacceptable risks to children and teens under 18 and should not be used by minors.

In practice: no simulated friendship or romance, no guilt or pressure to keep chatting, and plain explanations that it is a computer program that can be wrong.

4. Is it designed and filtered for their age?

UNESCO’s guidance on generative AI in education calls for an age limit for independent conversations with general AI platforms and an age-appropriate approach to their use. Tools for younger children need content screening on what goes in and what comes out.

In practice: a clear age range, safety filtering on every interaction, and no open access to general chatbots for young children.

5. Are adults kept in the loop without spying?

Parents and teachers need to see how a child is doing, but constant monitoring can stifle the freedom to try and fail. When AI is used to evaluate learning or monitor students during tests, the EU AI Act’s Annex III classes it as high-risk, with extra obligations.

In practice: share progress summaries, not transcripts of every word; let a human make any decision that affects a child’s education.

6. Can it be checked?

A framework only works if someone can verify it. Developers should publish their safety and privacy policies in plain language, explain how the system works at a level a parent can follow, and say who to contact when something goes wrong.

How does Clarity Kids apply these principles?

We build Clarity Kids, home of Studio Clarté, for children aged 6 to 12, and apply the framework to ourselves. A proprietary multi-stage safety filter screens every interaction. The depth of its answers can be configured. There are no ads and no tracking, and children’s data is never sold or used for advertising. Conversation data is deleted automatically after 90 days, as our privacy policy explains, and we publish our governance policies in full.

What should you ask before choosing an AI tool for a child?

1. Who is it designed for, and what age range?
2. Does it give hints and questions, or finished answers?
3. What data does it collect, how long is it kept, and is it ever used for advertising?
4. Does it pretend to be a friend?
5. What does it filter, and how can I report a problem?
6. Can I see a summary of how my child uses it?

For how to balance protection with teaching children about AI, see Should Children Use AI? Shielding vs AI Literacy.

Frequently asked questions

What makes an AI tool safe for children?

It is designed for their age, screens what it says, collects as little data as possible and deletes it on a schedule, does not advertise or profile, supports their thinking rather than doing it for them, and keeps parents and teachers involved.

What does UNICEF recommend for AI and children?

UNICEF’s Guidance on AI and children (version 3.0, December 2025) sets ten requirements, including ensuring safety, protecting children’s data and privacy, transparency and accountability, and supporting children’s best interests, development and well-being.

How long can an app keep a child’s data?

Under the US COPPA Rule, personal information from a child under 13 may be kept only as long as is reasonably necessary for the purpose it was collected for, never indefinitely, and the operator needs a written retention policy.

Should children use AI companion chatbots?

Common Sense Media advises that social AI companions should not be used by anyone under 18, after concluding in 2025 that they pose unacceptable risks to children and teens.

References

1. UNICEF (2025), Guidance on AI and children, version 3.0 — Ten requirements for child-centred AI, drawing on the Convention on the Rights of the Child; updated for AI companions and generative AI.
2. US Code of Federal Regulations, 16 CFR Part 312: Children's Online Privacy Protection Rule — Verifiable parental consent for children under 13; data kept only as long as reasonably necessary; written retention policy.
3. Bastani et al. (2025), Generative AI without guardrails can harm learning: Evidence from high school mathematics, PNAS — A tutor with safeguards that gave hints avoided the learning loss caused by an unrestricted chatbot.
4. Common Sense Media (2025), AI Companions Decoded — Social AI companions pose unacceptable risks to under-18s and should not be used by minors.
5. EU Artificial Intelligence Act, Annex III: High-Risk AI Systems — AI used to evaluate learning outcomes or monitor students during tests is high-risk.
6. UNESCO (2023), Guidance for generative AI in education and research — Recommends data privacy protection and an age limit for independent use of generative AI, with a human-centred, age-appropriate approach.