Safe-Child-LLM: A Developmental Benchmark for Evaluating LLM Safety in Child-AI Interactions

2025-06-16

AI Summary

Current LLM safety evaluations overlook developmental characteristics of minors, leading to critical gaps in protecting children and adolescents. Method: We introduce SG-Bench, the first age-specific safety benchmark for children (7–12 years) and adolescents (13–17 years), comprising 200 manually annotated, multi-stage adversarial prompts. We propose a cognitively grounded evaluation framework and a standardized 0–5 ethical refusal scale, and construct a red-teaming prompt library by integrating HarmBench, human annotation, cross-model evaluation across eight LLMs, and fine-grained safety scoring. Contribution/Results: Our systematic analysis reveals previously unreported safety blind spots in mainstream LLMs regarding child-sensitive topics—e.g., bodily privacy and peer pressure—and demonstrates their pervasive lack of robust, age-appropriate refusal capabilities. All datasets, annotations, and code are publicly released to support reproducible, developmentally informed LLM safety research.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Security and Privacy: Large-scale security measurements
• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

As Large Language Models (LLMs) increasingly power applications used by children and adolescents, ensuring safe and age-appropriate interactions has become an urgent ethical imperative. Despite progress in AI safety, current evaluations predominantly focus on adults, neglecting the unique vulnerabilities of minors engaging with generative AI. We introduce Safe-Child-LLM, a comprehensive benchmark and dataset for systematically assessing LLM safety across two developmental stages: children (7-12) and adolescents (13-17). Our framework includes a novel multi-part dataset of 200 adversarial prompts, curated from red-teaming corpora (e.g., SG-Bench, HarmBench), with human-annotated labels for jailbreak success and a standardized 0-5 ethical refusal scale. Evaluating leading LLMs -- including ChatGPT, Claude, Gemini, LLaMA, DeepSeek, Grok, Vicuna, and Mistral -- we uncover critical safety deficiencies in child-facing scenarios. This work highlights the need for community-driven benchmarks to protect young users in LLM interactions. To promote transparency and collaborative advancement in ethical AI development, we are publicly releasing both our benchmark datasets and evaluation codebase at https://github.com/The-Responsible-AI-Initiative/Safe_Child_LLM_Benchmark.git

Problem

• Evaluating LLM safety for child-AI interactions
• Addressing lack of minor-focused safety benchmarks
• Assessing age-appropriate responses across developmental stages

Innovation

• Introduces Safe-Child-LLM benchmark for child safety
• Uses 200 adversarial prompts with human annotations
• Evaluates multiple LLMs for ethical refusal scale