Guidance on AI and children · UNICEF
Source published: December 2025 · Version 3.0
Read the original at UNICEF InnocentiThe source in focus
UNICEF’s third edition of its guidance sets out ten requirements for child-centred AI. They address safety, privacy, fairness, accountability, inclusion and other rights considerations. The update also considers developments including AI companions, supply-chain risks and the use of AI in conflict.
The AIHI perspective
For organisations serving children or vulnerable communities, these questions belong in the opportunity brief. A promising feature is not a complete use case. Describe who could be affected, what information the workflow uses, what could go wrong and which person can stop or change the process.
Give review a practical shape
Bring programme, safeguarding, privacy and technical colleagues into the same conversation early. Ask how a person would challenge an output, how an error would be corrected and what the team would do if the system were unavailable. Use the relevant guidance and specialist advice to assess the actual deployment context.
Let the boundary improve the design
A smaller first workflow can be easier to explain, review and learn from. Start with a task where support is useful and responsibility remains clear. Document the evidence needed before expanding it, and involve the people who understand the service and its users.
Take one useful step.
- Describe who could be affected by the proposed workflow.
- Include the appropriate safeguarding and privacy reviewers.
- Define escalation and correction before the first live use.
Original AIHI commentary. Source authorship remains with the credited publication.
Explore your AI readiness
