This research introduces MinorBench, a specialized benchmark for assessing content-based risks that Large Language Models pose to children.

• Draws insights from a real-world deployment of an LLM chatbot in a middle school setting
• Proposes a comprehensive taxonomy of content-based risks specific to minors
• Provides concrete examples of how students use and sometimes misuse AI systems
• Addresses a critical gap in AI safety research focused on protecting younger users

For educators and EdTech developers, this research offers essential guidelines for safely integrating AI in learning environments while protecting children from inappropriate content, misinformation, and other unique risks.

MinorBench: A hand-built benchmark for content-based risks for children