The Children's AI Harm Registry

The Children's AI Harm Registry is a structured dataset documenting cases in which artificial intelligence systems caused or contributed to harm affecting people under the age of 18. To the maintainer's knowledge, it is the most comprehensive civilian-maintained dataset focused specifically on AI harms to children. Version 1.0 contains 484 documented cases spanning 2016 to 2026 across 44 named countries plus 40 multi-jurisdictional cases. Each case is classified across 21 dimensions including AI system, harm type, severity, mechanism, safeguard failure, and legal response. Cases carry one of two verification tiers: Verified (2+ independent sources, 31 cases) or Documented (single citable source, 453 cases). This honest evidentiary labeling is recorded per case. Upgrading documented cases to verified status is an active, ongoing priority for subsequent versions. The registry provides an empirical, citable evidence base for researchers, policymakers, child-safety organizations, journalists, and AI developers. It is designed to sit upstream of formal AI safety and evaluation work: where internal red-teaming generates synthetic failure modes, this registry catalogues documented real-world ones with attribution. Full methodology, inclusion and exclusion criteria, column dictionary, and known limitations are in the included README.