跳到正文
arXiv cs.AI· Zexuan Fei, Zhengnan Guo, Yiming Luo, Kaiyi Huang, Aimin Zhou, Bo Jiang, Fei Tan·· 4 小时前AI 评分34

EGS:面向教育 LLM 的互补安全框架

EGS: A Complementary Safety Framework for Educational LLMs

AI 导读

研究者提出教育护栏系统 EGS,一个轻量、与模型无关的运行时框架,通过查询审核、场景感知教学引导和回复审核,补充模型内在对齐。EGS 按风险等级分流输入,对灰色地带查询施以分类教学干预,并在输出前审计。在通用与教育专项基准上,EGS 提升了多种 LLM 的审核表现、拒答质量与越狱鲁棒性,代码与模型权重将公开。

正文 · 原文

View PDF HTML (experimental)

Abstract:Large Language Models (LLMs) offer strong potential for personalized education, but adolescent-facing deployment raises risks of toxic, biased, misleading, or developmentally inappropriate content. We argue that educational safety should move beyond binary refusal: student queries are often ambiguous or curiosity-driven, requiring contextual guidance rather than simple blocking. We introduce the Educational Guardrail System (EGS), a lightweight, model-agnostic runtime framework that complements intrinsic alignment through query moderation, scenario-aware pedagogical guidance, and response moderation. EGS triages inputs by risk level, redirects gray-zone queries with category-specific educational interventions, and audits outputs before delivery. Experiments across general and education-specific benchmarks show that EGS improves moderation performance, refusal quality, and jailbreak robustness across diverse LLMs. These findings suggest that defense-in-depth guardrails can strengthen safety while preserving the pedagogical utility required in educational LLMs. Our code and model weights will be publicly available to support community adoption.
Comments: 20 pages, 8 figures, 8 tables
Subjects: Computation and Language (cs.CL); Artificial Intelligence (cs.AI)
Cite as: arXiv:2611.22203 [cs.CL]
  (or arXiv:2611.22203v1 [cs.CL] for this version)
  https://doi.org/10.48550/arXiv.2611.22203

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Zexuan Fei [view email]
[v1] Tue, 17 Nov 2026 04:40:42 UTC (3,497 KB)

来源:arXiv cs.AI · arxiv.org