Understanding Generative AI Risks for Youth: A Taxonomy Based on Empirical Data

TL;DR

This study tackles the problem of emerging Generative AI risks for youth by building an empirically grounded risk taxonomy tailored to youth-GAI interactions. It triangulates three data sources—344 youth-GAI chat logs, 30,305 Reddit discussions, and 153 AI incidents—and employs inductive thematic analysis to derive six high-level risk categories and 84 specific risks, mapped to four interaction pathways. The key contributions include novel risk types (Mental Wellbeing and Behavioral/Social Developmental risks), a data-driven framework for moderation and intervention, and actionable guidance for developers, educators, caregivers, and policymakers to mitigate youth harms in GAI-enabled environments. The findings have practical significance for designing safer AI systems, informing youth safety policies, and guiding family-level decisions around GAI use among youth.

Abstract

Generative AI (GAI) is reshaping the way young users engage with technology. This study introduces a taxonomy of risks associated with youth-GAI interactions, derived from an analysis of 344 chat transcripts between youth and GAI chatbots, 30,305 Reddit discussions concerning youth engagement with these systems, and 153 documented AI-related incidents. We categorize risks into six overarching themes, identifying 84 specific risks, which we further align with four distinct interaction pathways. Our findings highlight emerging concerns, such as risks to mental wellbeing, behavioral and social development, and novel forms of toxicity, privacy breaches, and misuse/exploitation that are not fully addressed in existing frameworks on child online safety or AI risks. By systematically grounding our taxonomy in empirical data, this work offers a structured approach to aiding AI developers, educators, caregivers, and policymakers in comprehending and mitigating risks associated with youth-GAI interactions.

Figures (2)

• Figure 1: Overview of the Youth-AI Risk Taxonomy. The taxonomy consists of 84 unique low-level risk types, which are further categorized into 15 medium-level and 6 high-level risk types (and Other Risk). The sunburst plot visualizes the hierarchy, mapping high-level risk types (inner circle) to medium-level ones (outer ring). Low-level risks are numbered to align with their corresponding medium-level categories.
• Figure 2: This figure illustrates the four overarching typologies of harm and their connections to high-level risk types.