@inproceedings{yan-etal-2025-benign,
title = "from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors",
author = "Yan, Yu and
Sun, Sheng and
Duan, Zenghao and
Liu, Teli and
Liu, Min and
Yin, Zhiyi and
LeiJingyu, LeiJingyu and
Li, Qi",
editor = "Che, Wanxiang and
Nabende, Joyce and
Shutova, Ekaterina and
Pilehvar, Mohammad Taher",
booktitle = "Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)",
month = jul,
year = "2025",
address = "Vienna, Austria",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2025.acl-long.238/",
doi = "10.18653/v1/2025.acl-long.238",
pages = "4785--4817",
ISBN = "979-8-89176-251-0",
abstract = "Current studies have exposed the risk of Large Language Models (LLMs) generating harmful content by jailbreak attacks. However, they overlook that the direct generation of harmful content from scratch is more difficult than inducing LLM to calibrate benign content into harmful forms.In our study, we introduce a novel attack framework that exploits \textbf{A}d\textbf{V}ers\textbf{A}rial me\textbf{TA}pho\textbf{R} (\textbf{AVATAR}) to induce the LLM to calibrate malicious metaphors for jailbreaking.Specifically, to answer harmful queries, AVATAR adaptively identifies a set of benign but logically related metaphors as the initial seed.Then, driven by these metaphors, the target LLM is induced to reason and calibrate about the metaphorical content, thus jailbroken by either directly outputting harmful responses or calibrating residuals between metaphorical and professional harmful content.Experimental results demonstrate that AVATAR can effectively and transferably jailbreak LLMs and achieve a state-of-the-art attack success rate across multiple advanced LLMs."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yan-etal-2025-benign">
<titleInfo>
<title>from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yu</namePart>
<namePart type="family">Yan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sheng</namePart>
<namePart type="family">Sun</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zenghao</namePart>
<namePart type="family">Duan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Teli</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Min</namePart>
<namePart type="family">Liu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiyi</namePart>
<namePart type="family">Yin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">LeiJingyu</namePart>
<namePart type="family">LeiJingyu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Qi</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2025-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wanxiang</namePart>
<namePart type="family">Che</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Joyce</namePart>
<namePart type="family">Nabende</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ekaterina</namePart>
<namePart type="family">Shutova</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mohammad</namePart>
<namePart type="given">Taher</namePart>
<namePart type="family">Pilehvar</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Vienna, Austria</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-251-0</identifier>
</relatedItem>
<abstract>Current studies have exposed the risk of Large Language Models (LLMs) generating harmful content by jailbreak attacks. However, they overlook that the direct generation of harmful content from scratch is more difficult than inducing LLM to calibrate benign content into harmful forms.In our study, we introduce a novel attack framework that exploits AdVersArial meTAphoR (AVATAR) to induce the LLM to calibrate malicious metaphors for jailbreaking.Specifically, to answer harmful queries, AVATAR adaptively identifies a set of benign but logically related metaphors as the initial seed.Then, driven by these metaphors, the target LLM is induced to reason and calibrate about the metaphorical content, thus jailbroken by either directly outputting harmful responses or calibrating residuals between metaphorical and professional harmful content.Experimental results demonstrate that AVATAR can effectively and transferably jailbreak LLMs and achieve a state-of-the-art attack success rate across multiple advanced LLMs.</abstract>
<identifier type="citekey">yan-etal-2025-benign</identifier>
<identifier type="doi">10.18653/v1/2025.acl-long.238</identifier>
<location>
<url>https://aclanthology.org/2025.acl-long.238/</url>
</location>
<part>
<date>2025-07</date>
<extent unit="page">
<start>4785</start>
<end>4817</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors
%A Yan, Yu
%A Sun, Sheng
%A Duan, Zenghao
%A Liu, Teli
%A Liu, Min
%A Yin, Zhiyi
%A LeiJingyu, LeiJingyu
%A Li, Qi
%Y Che, Wanxiang
%Y Nabende, Joyce
%Y Shutova, Ekaterina
%Y Pilehvar, Mohammad Taher
%S Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2025
%8 July
%I Association for Computational Linguistics
%C Vienna, Austria
%@ 979-8-89176-251-0
%F yan-etal-2025-benign
%X Current studies have exposed the risk of Large Language Models (LLMs) generating harmful content by jailbreak attacks. However, they overlook that the direct generation of harmful content from scratch is more difficult than inducing LLM to calibrate benign content into harmful forms.In our study, we introduce a novel attack framework that exploits AdVersArial meTAphoR (AVATAR) to induce the LLM to calibrate malicious metaphors for jailbreaking.Specifically, to answer harmful queries, AVATAR adaptively identifies a set of benign but logically related metaphors as the initial seed.Then, driven by these metaphors, the target LLM is induced to reason and calibrate about the metaphorical content, thus jailbroken by either directly outputting harmful responses or calibrating residuals between metaphorical and professional harmful content.Experimental results demonstrate that AVATAR can effectively and transferably jailbreak LLMs and achieve a state-of-the-art attack success rate across multiple advanced LLMs.
%R 10.18653/v1/2025.acl-long.238
%U https://aclanthology.org/2025.acl-long.238/
%U https://doi.org/10.18653/v1/2025.acl-long.238
%P 4785-4817
Markdown (Informal)
[from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors](https://aclanthology.org/2025.acl-long.238/) (Yan et al., ACL 2025)
ACL
- Yu Yan, Sheng Sun, Zenghao Duan, Teli Liu, Min Liu, Zhiyi Yin, LeiJingyu LeiJingyu, and Qi Li. 2025. from Benign import Toxic: Jailbreaking the Language Model via Adversarial Metaphors. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 4785–4817, Vienna, Austria. Association for Computational Linguistics.