Before AI reaches children, should it have to prove it is safe?
An international coalition co-chaired by France, Spain, Kenya and Australia has launched a 2026 to 2027 roadmap for children’s rights and protection in the age of artificial intelligence. The roadmap brings participating governments and United Nations entities together around safety and privacy by design, accountability, AI literacy, independent research and stronger responses to AI-generated sexual abuse material and deepfakes.
What it could improve
The roadmap places responsibility on those who design, deploy and govern AI systems, rather than expecting children and families to manage every risk themselves. Its longer-term exploration of pre-authorisation approaches for AI systems claiming benefits for children is especially important. It asks whether such systems should demonstrate safety, effectiveness and respect for children’s rights before they reach young users.
What remains unresolved
The roadmap is a voluntary international coordination framework. Its value will depend on whether common principles become measurable requirements that operate consistently across countries, networks and services.
- Participating governments still need shared methods for evaluating developmental, privacy, discrimination, manipulation and wellbeing risks before and after deployment.
- Independent researchers need meaningful access to systems, data and evidence without weakening children’s privacy or platform security.
- AI literacy is valuable, but it should not transfer responsibility for unsafe systems back to children, parents or teachers.
- AI services may not know when a user is a child, while repeated provider-by-provider age checks can expose more personal information than is necessary.
- Voluntary commitments need transparent benchmarks, reporting and routes toward enforceable duties if harmful practices continue.
- Protection must extend beyond well-known chatbots to embedded assistants, games, learning tools and new AI services that children encounter across the Internet.
The Child-Friendly Internet perspective
The roadmap strongly supports the Child-Friendly Internet principle that technology should adapt to children. Safety and privacy should be demonstrated before an AI system is promoted for children, then monitored throughout its life. A privacy-preserving assurance layer could allow participating AI services to receive only the developmental context needed for an appropriate response, without receiving the child’s identity or creating a shared tracking identifier.
- Develop pre-market and continuing assessments that test real effects on children, not only technical performance.
- Require age-appropriate responses, safer defaults and clear limits on persuasive or emotionally dependent interaction.
- Use purpose-limited assurance so services learn only what they need to provide an appropriate experience.
- Support children’s participation, accessibility, beneficial use and growing autonomy alongside protection.
- Create interoperable standards so protection does not stop when a child moves between AI services, devices or networks.
The central question
How can voluntary international principles become consistent, enforceable protection without requiring every AI service to identify the child again?