@inproceedings{yang-etal-2026-visual,
title = "Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images",
author = "Yang, Qishun and
Yang, Shu and
Hu, Lijie and
Wang, Di",
editor = "Liakata, Maria and
Moreira, Viviane P. and
Zhang, Jiajun and
Jurgens, David",
booktitle = "Proceedings of the 64th Annual Meeting of the {A}ssociation for {C}omputational {L}inguistics (Volume 1: Long Papers)",
month = jul,
year = "2026",
address = "San Diego, California, United States",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.acl-long.490/",
pages = "10698--10718",
ISBN = "979-8-89176-390-6",
abstract = "Multimodal large language models (MLLMs) face safety misalignment where visual inputs enable harmful outputs. Existing methods require explicit safety labels or contrastive data, yet threat-related concepts are concrete and visually depictable, while safety concepts like helpfulness are abstract and lack visual referents. Inspired by self-fulfilling mechanism underlying emergent misalignment, we propose Visual Self-Fulfilling Alignment (VSFA). VSFA fine-tunes vision-language models (VLMs) on neutral VQA tasks constructed around threat-related images, without any safety labels. Through repeated exposure to threat-related visual content, models internalize implicit semantics of vigilance and caution, shaping safety-oriented personas. Experiments across multiple VLMs and safety benchmarks demonstrate that VSFA reduces attack success rate, improves response quality, and mitigates over-refusal while preserving general capabilities. Our work extends self-fulfilling mechanism from text to visual modalities, offering a label-free approach to VLMs alignment."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="yang-etal-2026-visual">
<titleInfo>
<title>Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images</title>
</titleInfo>
<name type="personal">
<namePart type="given">Qishun</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shu</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Lijie</namePart>
<namePart type="family">Hu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Di</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-07</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)</title>
</titleInfo>
<name type="personal">
<namePart type="given">Maria</namePart>
<namePart type="family">Liakata</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Viviane</namePart>
<namePart type="given">P</namePart>
<namePart type="family">Moreira</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiajun</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">David</namePart>
<namePart type="family">Jurgens</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">San Diego, California, United States</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-390-6</identifier>
</relatedItem>
<abstract>Multimodal large language models (MLLMs) face safety misalignment where visual inputs enable harmful outputs. Existing methods require explicit safety labels or contrastive data, yet threat-related concepts are concrete and visually depictable, while safety concepts like helpfulness are abstract and lack visual referents. Inspired by self-fulfilling mechanism underlying emergent misalignment, we propose Visual Self-Fulfilling Alignment (VSFA). VSFA fine-tunes vision-language models (VLMs) on neutral VQA tasks constructed around threat-related images, without any safety labels. Through repeated exposure to threat-related visual content, models internalize implicit semantics of vigilance and caution, shaping safety-oriented personas. Experiments across multiple VLMs and safety benchmarks demonstrate that VSFA reduces attack success rate, improves response quality, and mitigates over-refusal while preserving general capabilities. Our work extends self-fulfilling mechanism from text to visual modalities, offering a label-free approach to VLMs alignment.</abstract>
<identifier type="citekey">yang-etal-2026-visual</identifier>
<location>
<url>https://aclanthology.org/2026.acl-long.490/</url>
</location>
<part>
<date>2026-07</date>
<extent unit="page">
<start>10698</start>
<end>10718</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images
%A Yang, Qishun
%A Yang, Shu
%A Hu, Lijie
%A Wang, Di
%Y Liakata, Maria
%Y Moreira, Viviane P.
%Y Zhang, Jiajun
%Y Jurgens, David
%S Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)
%D 2026
%8 July
%I Association for Computational Linguistics
%C San Diego, California, United States
%@ 979-8-89176-390-6
%F yang-etal-2026-visual
%X Multimodal large language models (MLLMs) face safety misalignment where visual inputs enable harmful outputs. Existing methods require explicit safety labels or contrastive data, yet threat-related concepts are concrete and visually depictable, while safety concepts like helpfulness are abstract and lack visual referents. Inspired by self-fulfilling mechanism underlying emergent misalignment, we propose Visual Self-Fulfilling Alignment (VSFA). VSFA fine-tunes vision-language models (VLMs) on neutral VQA tasks constructed around threat-related images, without any safety labels. Through repeated exposure to threat-related visual content, models internalize implicit semantics of vigilance and caution, shaping safety-oriented personas. Experiments across multiple VLMs and safety benchmarks demonstrate that VSFA reduces attack success rate, improves response quality, and mitigates over-refusal while preserving general capabilities. Our work extends self-fulfilling mechanism from text to visual modalities, offering a label-free approach to VLMs alignment.
%U https://aclanthology.org/2026.acl-long.490/
%P 10698-10718
Markdown (Informal)
[Visual Self-Fulfilling Alignment: Shaping Safety-Oriented Personas via Threat-Related Images](https://aclanthology.org/2026.acl-long.490/) (Yang et al., ACL 2026)
ACL