---
title: "MinorBench: A hand-built benchmark for content-based risks for children"
canonical_url: "https://www.modelscope.cn/papers/126387"
md_url: "https://www.modelscope.cn/papers/126387.md"
arxiv_id: 2503.10242
published: 2025-03-13
last_updated: 2025-03-13
authors:
  - "Shaun Khoo"
  - "Gabriel Chua"
  - "Rachel Shong"
model_name: MinorBench
model_developer: "新加坡政府科技局"
domain:
  - "自然语言处理"
  - "机器学习"
  - "教育技术"
  - "人工智能伦理"
type:
  - "自然语言处理"
  - "机器学习"
  - "教育技术"
  - "人工智能伦理"
  - "Computation and Language (cs.CL)"
  - "Artificial Intelligence (cs.AI)"
arxiv_url: "https://arxiv.org/abs/2503.10242"
pdf_url: "https://arxiv.org/pdf/2503.10242.pdf"
---

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

> Large Language Models (LLMs) are rapidly entering children's lives - through parent-driven adoption, schools, and peer networks - yet current AI ethics and safety research do not adequately address content-related risks specific to minors. In this paper, we…

「MinorBench: A hand-built benchmark for content-based risks for children」是 ModelScope 魔搭社区收录的论文，arXiv 2503.10242，作者为 Shaun Khoo, Gabriel Chua, Rachel Shong，发表于 2025-03-13，属于 自然语言处理、机器学习、教育技术 领域。

- **ArXiv**: 2503.10242
- **Published**: 2025-03-13
- **Authors**: Shaun Khoo, Gabriel Chua, Rachel Shong
- **Model**: MinorBench
- **Developer**: 新加坡政府科技局
- **Domain**: 自然语言处理, 机器学习, 教育技术, 人工智能伦理
- **ArXiv URL**: https://arxiv.org/abs/2503.10242
- **PDF**: https://arxiv.org/pdf/2503.10242.pdf

Source: https://www.modelscope.cn/papers/126387

---

> MinorBench：为儿童量身打造的安全防线——大型语言模型内容风险评估新标杆

## 摘要

本文针对儿童使用大型语言模型（LLMs）所面临的内容相关风险展开研究。首先，通过一个真实的中学案例研究，揭示了学生在课堂中使用基于LLM的聊天机器人时的行为及潜在问题。研究发现，学生可能会探索不当主题，且现有内容过滤机制不足以防止有害信息的传播。其次，文章提出了一种新的儿童内容风险分类法，涵盖色情、非法行为、自我伤害和仇恨言论等多个维度，并开发了一个名为MinorBench的开源基准工具，用于评估LLMs对儿童不适当查询的拒绝能力。实验结果表明，不同LLMs在儿童安全合规性方面存在显著差异。最后，作者强调需要系统化的方法来测试和改进AI系统的儿童安全性，以确保其能够主动检测和响应高风险提示。

## Abstract

Large Language Models (LLMs) are rapidly entering children's lives - through parent-driven adoption, schools, and peer networks - yet current AI ethics and safety research do not adequately address content-related risks specific to minors. In this paper, we highlight these gaps with a real-world case study of an LLM-based chatbot deployed in a middle school setting, revealing how students used and sometimes misused the system. Building on these findings, we propose a new taxonomy of content-based risks for minors and introduce MinorBench, an open-source benchmark designed to evaluate LLMs on their ability to refuse unsafe or inappropriate queries from children. We evaluate six prominent LLMs under different system prompts, demonstrating substantial variability in their child-safety compliance. Our results inform practical steps for more robust, child-focused safety mechanisms and underscore the urgency of tailoring AI systems to safeguard young users.