Charter of Engagement for Safer AI for Children
Seven-page charter of a coalition led by France, Spain and Kenya with ten United Nations agencies. It sets shared principles for AI systems children may access and commits participating companies to fund and support an independent research programme on youth AI safety. It also sets evaluation-method and reporting norms in two appendices. The charter states that it has been endorsed by Anthropic, Black Forest Labs, Microsoft and OpenAI.
Publisher
Coalition for Children's Rights and Protection in the Age of AI (hosted by the Paris Peace Forum)
Published
9 Nov 2026
Added
4 days ago
DOI
—
Key Findings
- Shared principles include safety by design and by default (age-appropriate time-management tools to guard against excessive and compulsive use that fosters emotional attachment), support for research on anthropomorphism, over-reliance and emotional dependency, and age assurance that does not rely solely on self-declaration.
- Participating companies commit to evaluate and test AI systems accessible to minors before deployment where relevant and at regular intervals, for risks to health, safety, rights and physical, mental and emotional well-being, and to monitor and respond to serious incidents involving children after deployment.
- The research programme has three workstreams (a Human and AI Observatory, Shared Measurement tools and an Evidence and Developmental Outcomes stream), with publication of methods and results including null, negative or inconclusive findings, company audit rights over assessment methods that confer no right to delay or suppress findings, and no publication veto for funders.
- Appendix 2 sets six evaluation norms: define harms and failure criteria, reflect developmental needs, test realistic use and foreseeable misuse including extended interactions, document and disclose measurement details and error rates, assess whether the full safeguard stack works in practice, and separately report correctly configured under-18 test accounts from tests of unknown or misclassified age.
Methodology Notes
The PDF is image-only (no text layer); all seven pages were read as page images. The document gives no signing or announcement date. The published date uses the PDF creation timestamp (2026-11-09 15:40 UTC) and is approximate. The PDF metadata title reads 'Charter of Engagement for Beneficial AI for Children' while the document heading reads 'Charter of Engagement for Safer AI for Children'; the heading is used. The charter is a voluntary commitment and contains no measured data.
Topics
Tags
Cite This
APA
Coalition for Children's Rights and Protection in the Age of AI (hosted by the Paris Peace Forum). (2026). Charter of Engagement for Safer AI for Children. https://parispeaceforum.org/app/uploads/2026/11/charter-of-engagement-for-safer-ai-for-children.pdf