@inproceedings{wang-etal-2026-gpc,
title = "{GPC}-{E}4: Generalized Consistency and Entropic Evidence for Backdoor Detection in Vision Language Models",
author = "Wang, Xuan and
Yan, Xuehu and
Lu, Yuliang and
Ye, Kejiang and
Xu, Cheng-zhong and
Gao, Xitong",
editor = "Zhang, Yue and
Xu, Ruifeng and
Purwarianti, Ayu and
Rong, Tong and
Suzuki, Jun",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {IJCNLP}-{AACL} 2026",
month = nov,
year = "2026",
address = "Hengqin, China",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-aacl.176/",
pages = "3507--3522",
ISBN = "979-8-89176-444-6",
abstract = "Vision-language models (VLMs) are increasingly used through black-box, open-ended interfaces, where input-level backdoor screening must work without model parameters, gradients, logits, or training data. We present GPC-E4, a detector for single image-prompt queries that probes a VLM with mild, semantics-preserving visual and textual perturbations. The method measures whether the output behavior contracts unusually under these probes: GPC captures dominant-mode or semantic consistency, while E4 summarizes predictive entropy, augmentation mutual information, minimal collapse, and local slope along perturbation paths. A fixed robust normalization combines these signals and can operate without a known-clean calibration set. Across COCO, Flickr30k, Caltech101, and MMBench-style evaluations with LLaVA-1.5, MiniGPT-4, CLIP, CoOp, and Qwen3-VL, GPC-E4 outperforms STRIP, SCALE, TeCo, and BDet in both closed-set and open-ended settings. Additional analyses under alternative stressors, prompt and distribution shifts, zero clean calibration, VLM-native attacks, and detector-aware attacks characterize when the contraction signal is reliable and when it weakens."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="wang-etal-2026-gpc">
<titleInfo>
<title>GPC-E4: Generalized Consistency and Entropic Evidence for Backdoor Detection in Vision Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xuan</namePart>
<namePart type="family">Wang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xuehu</namePart>
<namePart type="family">Yan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yuliang</namePart>
<namePart type="family">Lu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Kejiang</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Cheng-zhong</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xitong</namePart>
<namePart type="family">Gao</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-11</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yue</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruifeng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ayu</namePart>
<namePart type="family">Purwarianti</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tong</namePart>
<namePart type="family">Rong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jun</namePart>
<namePart type="family">Suzuki</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Hengqin, China</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-444-6</identifier>
</relatedItem>
<abstract>Vision-language models (VLMs) are increasingly used through black-box, open-ended interfaces, where input-level backdoor screening must work without model parameters, gradients, logits, or training data. We present GPC-E4, a detector for single image-prompt queries that probes a VLM with mild, semantics-preserving visual and textual perturbations. The method measures whether the output behavior contracts unusually under these probes: GPC captures dominant-mode or semantic consistency, while E4 summarizes predictive entropy, augmentation mutual information, minimal collapse, and local slope along perturbation paths. A fixed robust normalization combines these signals and can operate without a known-clean calibration set. Across COCO, Flickr30k, Caltech101, and MMBench-style evaluations with LLaVA-1.5, MiniGPT-4, CLIP, CoOp, and Qwen3-VL, GPC-E4 outperforms STRIP, SCALE, TeCo, and BDet in both closed-set and open-ended settings. Additional analyses under alternative stressors, prompt and distribution shifts, zero clean calibration, VLM-native attacks, and detector-aware attacks characterize when the contraction signal is reliable and when it weakens.</abstract>
<identifier type="citekey">wang-etal-2026-gpc</identifier>
<location>
<url>https://aclanthology.org/2026.findings-aacl.176/</url>
</location>
<part>
<date>2026-11</date>
<extent unit="page">
<start>3507</start>
<end>3522</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T GPC-E4: Generalized Consistency and Entropic Evidence for Backdoor Detection in Vision Language Models
%A Wang, Xuan
%A Yan, Xuehu
%A Lu, Yuliang
%A Ye, Kejiang
%A Xu, Cheng-zhong
%A Gao, Xitong
%Y Zhang, Yue
%Y Xu, Ruifeng
%Y Purwarianti, Ayu
%Y Rong, Tong
%Y Suzuki, Jun
%S Findings of the Association for Computational Linguistics: IJCNLP-AACL 2026
%D 2026
%8 November
%I Association for Computational Linguistics
%C Hengqin, China
%@ 979-8-89176-444-6
%F wang-etal-2026-gpc
%X Vision-language models (VLMs) are increasingly used through black-box, open-ended interfaces, where input-level backdoor screening must work without model parameters, gradients, logits, or training data. We present GPC-E4, a detector for single image-prompt queries that probes a VLM with mild, semantics-preserving visual and textual perturbations. The method measures whether the output behavior contracts unusually under these probes: GPC captures dominant-mode or semantic consistency, while E4 summarizes predictive entropy, augmentation mutual information, minimal collapse, and local slope along perturbation paths. A fixed robust normalization combines these signals and can operate without a known-clean calibration set. Across COCO, Flickr30k, Caltech101, and MMBench-style evaluations with LLaVA-1.5, MiniGPT-4, CLIP, CoOp, and Qwen3-VL, GPC-E4 outperforms STRIP, SCALE, TeCo, and BDet in both closed-set and open-ended settings. Additional analyses under alternative stressors, prompt and distribution shifts, zero clean calibration, VLM-native attacks, and detector-aware attacks characterize when the contraction signal is reliable and when it weakens.
%U https://aclanthology.org/2026.findings-aacl.176/
%P 3507-3522
Markdown (Informal)
[GPC-E4: Generalized Consistency and Entropic Evidence for Backdoor Detection in Vision Language Models](https://aclanthology.org/2026.findings-aacl.176/) (Wang et al., Findings 2026)
ACL