@inproceedings{jia-etal-2026-castle,
title = "{CASTLE}: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models",
author = "Jia, Rui and
Lan, Ruiyi and
Liu, Fengrui and
Dai, Zhongxiang and
Jiang, Bo and
Shao, Jing and
Chen, Jingyuan and
Xu, Guandong and
Wu, Fei and
Zhang, Min",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.1008/",
pages = "23188--23222",
ISBN = "979-8-89176-459-0",
abstract = "Large language models (LLMs) have advanced the development of personalized learning in education. However, their inherent generation mechanisms often produce homogeneous responses to identical prompts. This ``one-size-fits-all'' mechanism overlooks the substantial heterogeneity in students' cognitive and psychological characteristics, thereby posing potential safety risks to vulnerable groups. Existing safety evaluations primarily rely on context-independent metrics such as factual accuracy, bias, or toxicity, which fail to capture the divergent harms that the same response might cause across different student attributes. To address this gap, we propose the concept of ``Student-Tailored Personalized Safety'' and construct CASTLE based on educational theories. This benchmark covers 15 educational safety risks and 14 student attributes, comprising 92,908 bilingual scenarios. We further design three evaluation metrics: Risk Sensitivity, measuring the model{'}s ability to detect risks; Emotional Empathy, evaluating the model{'}s capacity to recognize student states; and Student Alignment, assessing the match between model responses and student attributes. Experiments on 18 LLMs demonstrate that CASTLE poses a significant challenge: all models scored below an average safety score of 2.5 out of 5, indicating substantial deficiencies in personalized safety assurance. The code is available at https://github.com/ECNU-RAIL/CASTLE-EMNLP2026."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jia-etal-2026-castle">
<titleInfo>
<title>CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Rui</namePart>
<namePart type="family">Jia</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruiyi</namePart>
<namePart type="family">Lan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fengrui</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhongxiang</namePart>
<namePart type="family">Dai</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Bo</namePart>
<namePart type="family">Jiang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jing</namePart>
<namePart type="family">Shao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jingyuan</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guandong</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Fei</namePart>
<namePart type="family">Wu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Large language models (LLMs) have advanced the development of personalized learning in education. However, their inherent generation mechanisms often produce homogeneous responses to identical prompts. This “one-size-fits-all” mechanism overlooks the substantial heterogeneity in students’ cognitive and psychological characteristics, thereby posing potential safety risks to vulnerable groups. Existing safety evaluations primarily rely on context-independent metrics such as factual accuracy, bias, or toxicity, which fail to capture the divergent harms that the same response might cause across different student attributes. To address this gap, we propose the concept of “Student-Tailored Personalized Safety” and construct CASTLE based on educational theories. This benchmark covers 15 educational safety risks and 14 student attributes, comprising 92,908 bilingual scenarios. We further design three evaluation metrics: Risk Sensitivity, measuring the model’s ability to detect risks; Emotional Empathy, evaluating the model’s capacity to recognize student states; and Student Alignment, assessing the match between model responses and student attributes. Experiments on 18 LLMs demonstrate that CASTLE poses a significant challenge: all models scored below an average safety score of 2.5 out of 5, indicating substantial deficiencies in personalized safety assurance. The code is available at https://github.com/ECNU-RAIL/CASTLE-EMNLP2026.</abstract>
<identifier type="citekey">jia-etal-2026-castle</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1008/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>23188</start>
<end>23222</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models
%A Jia, Rui
%A Lan, Ruiyi
%A Liu, Fengrui
%A Dai, Zhongxiang
%A Jiang, Bo
%A Shao, Jing
%A Chen, Jingyuan
%A Xu, Guandong
%A Wu, Fei
%A Zhang, Min
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F jia-etal-2026-castle
%X Large language models (LLMs) have advanced the development of personalized learning in education. However, their inherent generation mechanisms often produce homogeneous responses to identical prompts. This “one-size-fits-all” mechanism overlooks the substantial heterogeneity in students’ cognitive and psychological characteristics, thereby posing potential safety risks to vulnerable groups. Existing safety evaluations primarily rely on context-independent metrics such as factual accuracy, bias, or toxicity, which fail to capture the divergent harms that the same response might cause across different student attributes. To address this gap, we propose the concept of “Student-Tailored Personalized Safety” and construct CASTLE based on educational theories. This benchmark covers 15 educational safety risks and 14 student attributes, comprising 92,908 bilingual scenarios. We further design three evaluation metrics: Risk Sensitivity, measuring the model’s ability to detect risks; Emotional Empathy, evaluating the model’s capacity to recognize student states; and Student Alignment, assessing the match between model responses and student attributes. Experiments on 18 LLMs demonstrate that CASTLE poses a significant challenge: all models scored below an average safety score of 2.5 out of 5, indicating substantial deficiencies in personalized safety assurance. The code is available at https://github.com/ECNU-RAIL/CASTLE-EMNLP2026.
%U https://aclanthology.org/2026.emnlp-main.1008/
%P 23188-23222
Markdown (Informal)
[CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models](https://aclanthology.org/2026.emnlp-main.1008/) (Jia et al., EMNLP 2026)
ACL
- Rui Jia, Ruiyi Lan, Fengrui Liu, Zhongxiang Dai, Bo Jiang, Jing Shao, Jingyuan Chen, Guandong Xu, Fei Wu, and Min Zhang. 2026. CASTLE: A Comprehensive Benchmark for Evaluating Student-Tailored Personalized Safety in Large Language Models. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 23188–23222, Budapest, Hungary. Association for Computational Linguistics.