@inproceedings{li-etal-2026-text,
title = "Text-Anchored Semantic Perturbations for Transferable Jailbreak Attacks on Multimodal Large Language Models",
author = "Li, Wenyun and
Cao, Guiping and
Lan, Xiangyuan and
Zhang, Zheng",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.165/",
pages = "3500--3515",
ISBN = "979-8-89176-462-0",
abstract = "Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language interaction, yet their safety alignment remains vulnerable to jailbreak attacks. A key challenge is that safety behavior learned in the textual space does not reliably transfer to fused cross-modal representations, leaving multimodal inputs exploitable through latent semantic cues. We propose {\_}{\_}Text-Anchored Semantic Perturbation Attack{\_}{\_} (TA-SPA), a black-box jailbreak framework that optimizes transferable perturbations in a text-anchored semantic space. TA-SPA integrates {\_}{\_}Text-Anchored Semantic Factorization{\_}{\_} (TASF), which disentangles cross-modal semantic factors from modality-specific residuals, with {\_}{\_}Semantic-Preserving Augmentation{\_}{\_} (SPA), which diversifies harmful target anchors while preserving semantic consistency. Extensive experiments show that TA-SPA consistently outperforms prior jailbreak attacks, transfers to commercial MLLMs, and remains effective under representative defenses, underscoring the need for representation-level safety alignment beyond input-level filtering."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="li-etal-2026-text">
<titleInfo>
<title>Text-Anchored Semantic Perturbations for Transferable Jailbreak Attacks on Multimodal Large Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Wenyun</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Guiping</namePart>
<namePart type="family">Cao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiangyuan</namePart>
<namePart type="family">Lan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zheng</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language interaction, yet their safety alignment remains vulnerable to jailbreak attacks. A key challenge is that safety behavior learned in the textual space does not reliably transfer to fused cross-modal representations, leaving multimodal inputs exploitable through latent semantic cues. We propose __Text-Anchored Semantic Perturbation Attack__ (TA-SPA), a black-box jailbreak framework that optimizes transferable perturbations in a text-anchored semantic space. TA-SPA integrates __Text-Anchored Semantic Factorization__ (TASF), which disentangles cross-modal semantic factors from modality-specific residuals, with __Semantic-Preserving Augmentation__ (SPA), which diversifies harmful target anchors while preserving semantic consistency. Extensive experiments show that TA-SPA consistently outperforms prior jailbreak attacks, transfers to commercial MLLMs, and remains effective under representative defenses, underscoring the need for representation-level safety alignment beyond input-level filtering.</abstract>
<identifier type="citekey">li-etal-2026-text</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.165/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>3500</start>
<end>3515</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Text-Anchored Semantic Perturbations for Transferable Jailbreak Attacks on Multimodal Large Language Models
%A Li, Wenyun
%A Cao, Guiping
%A Lan, Xiangyuan
%A Zhang, Zheng
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F li-etal-2026-text
%X Multimodal Large Language Models (MLLMs) have achieved remarkable progress in vision-language interaction, yet their safety alignment remains vulnerable to jailbreak attacks. A key challenge is that safety behavior learned in the textual space does not reliably transfer to fused cross-modal representations, leaving multimodal inputs exploitable through latent semantic cues. We propose __Text-Anchored Semantic Perturbation Attack__ (TA-SPA), a black-box jailbreak framework that optimizes transferable perturbations in a text-anchored semantic space. TA-SPA integrates __Text-Anchored Semantic Factorization__ (TASF), which disentangles cross-modal semantic factors from modality-specific residuals, with __Semantic-Preserving Augmentation__ (SPA), which diversifies harmful target anchors while preserving semantic consistency. Extensive experiments show that TA-SPA consistently outperforms prior jailbreak attacks, transfers to commercial MLLMs, and remains effective under representative defenses, underscoring the need for representation-level safety alignment beyond input-level filtering.
%U https://aclanthology.org/2026.findings-emnlp.165/
%P 3500-3515
Markdown (Informal)
[Text-Anchored Semantic Perturbations for Transferable Jailbreak Attacks on Multimodal Large Language Models](https://aclanthology.org/2026.findings-emnlp.165/) (Li et al., Findings 2026)
ACL