About AABoardGovernancePeopleNewsContact
AI Assessment Board (AABoard)Advancing AI literacy through evidence and standardsSearch registry
Back to News

News

Yorba Linda AI Storytime pilot informs research on child-facing generative AI safety

Nov. 2025

AABoard has linked AAB-PILOT-2025-LL-001: Yorba Linda Public Library AI Storytime, a public-library AI storytelling pilot with elementary-age learners in Yorba Linda, Southern California, to a related research preprint on child-facing generative AI storytelling safety.

The manuscript, When a Safe Prompt Produces an Unsafe Feeling: The Context-Ambiguity Gap in Child-Facing Generative AI Storytelling, analyzes an implementation issue identified during the pilot: a benign child prompt and positive story text can still produce an AI-generated image that a young child may interpret as emotionally unsafe.

Related research publication

The pilot record now includes a related publications section that points readers to the preprint and preserves the connection between the implementation record, the incident analysis, and the resulting design recommendations.

Key contributions and findings

  • The paper names the context-ambiguity gap: an AI output can be semantically related to a safe prompt while lacking enough narrative, emotional, or adult-mediated context for a young child to interpret it safely.
  • It argues that prompt-level safety is necessary but incomplete for child-facing multimodal AI. A cheerful prompt can still lead to a visually ambiguous or unsettling image.
  • The analysis shows that static AI-generated images can remove the recovery cues that children's media normally provides through dialogue, pacing, music, facial expression, and narrative repair.
  • The paper proposes Developmental Interpretation Safety as a design requirement for child-facing AI storytelling: outputs should be emotionally safe, narratively coherent, and recoverable when the system makes a mistake.
  • It recommends practical safeguards including constrained prompting, adult preview, visual ambiguity checks, positive-ending defaults, immediate regeneration, and facilitator repair language.

Why the pilot matters

Yorba Linda Public Library AI Storytime documents a one-time 90-minute informal learning workshop using a generative AI storytelling application with young learners. The pilot record captures the setting, audience, privacy-preserving implementation choices, facilitator mediation, prompt scaffolding, observed implementation constraints, and the need for clearer procedures when a generated image is confusing or emotionally ambiguous.

For AABoard, this is the type of evidence link that makes a pilot registry useful: local implementation is documented first, then connected to analysis that can be cited, reviewed, compared, and reused by educators, researchers, standards contributors, and designers of child-facing AI literacy activities.

Practical implications for future events

The paper turns the pilot issue into a concrete event protocol. For young children, facilitators should use closed-choice handouts, include clear positive-ending language, preview generated images before showing them to children, regenerate images that appear injured or lifeless, and use calm repair scripts such as: "The AI got confused. Let's help it make a better picture."

Evidence boundary

AABoard treats the pilot as implementation evidence, not as a controlled study or certification claim. The manuscript is a design-case analysis based on one documented incident and local event artifacts. Its value is not a prevalence claim; its value is identifying a repeatable design risk and a practical safeguard framework for future child-facing AI literacy work.