Adversarial Attacks on Vision-Language Multimodal Systems: A Survey

With the boom of artificial intelligence, multimodal deep learning, especially large language models (LLMs), has advanced remarkably and found wide applications. However, adversarial security issues hamper further development on LLMs. This paper comprehensively reviews adversarial attacks on vision-language multimodal systems (VLMS), covering key aspects such as pretrained models, large language models, adversarial attack methodologies, defense strategies, benchmark datasets, downstream tasks, and evaluation metrics. It first details multimodal deep-learning key technologies and the features of relevant models. Then, it deeply analyzes the adversarial vulnerabilities of VLMS. Next, it systematically reviews various adversarial attack methods and defend strategies in multimodal systems, covering task-specific attacks and attacks on different models, with a detailed analysis of their principles, implementations, and effects. Finally, it looks ahead to the future of multimodal adversarial attacks, suggesting research directions like improving attack universality and compliance considerations. This review offers technical references, promotes research on multimodal adversarial attacks, and helps enhance VLMS security for reliable real-world applications. The latest papers on VLMS adversarial attacks are continuously collected in https://github.com/chilljudaoren/A-Survey-of-Adversarial-Attack.

Authors

Institutions

Publication Details

Journal
ACM Transactions on Privacy and Security
Published
2026-11-21
DOI
https://doi.org/10.1145/3871040
Primary Topic
Adversarial Robustness in Machine Learning
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Adversarial Attacks on Vision-Language Multimodal Systems: A Survey

Qiqi Bao, Hegui Zhu, Chenhao Lin, Yaguan Qian et al.
ACM Transactions on Privacy and Security
Adversarial Robustness in Machine Learning
article

Adversarial Attacks on Vision-Language Multimodal Systems: A Survey

Qiqi Bao, Hegui Zhu, Chenhao Lin, Yaguan Qian, Shouling Ji, Zhaoquan Gu, Yan Gan, Jia-Li Yin, Zhihao Chen, Shuo Wang, Wei Wang, Bin Wang
article en

Abstract

With the boom of artificial intelligence, multimodal deep learning, especially large language models (LLMs), has advanced remarkably and found wide applications. However, adversarial security issues hamper further development on LLMs. This paper comprehensively reviews adversarial attacks on vision-language multimodal systems (VLMS), covering key aspects such as pretrained models, large language models, adversarial attack methodologies, defense strategies, benchmark datasets, downstream tasks, and evaluation metrics. It first details multimodal deep-learning key technologies and the features of relevant models. Then, it deeply analyzes the adversarial vulnerabilities of VLMS. Next, it systematically reviews various adversarial attack methods and defend strategies in multimodal systems, covering task-specific attacks and attacks on different models, with a detailed analysis of their principles, implementations, and effects. Finally, it looks ahead to the future of multimodal adversarial attacks, suggesting research directions like improving attack universality and compliance considerations. This review offers technical references, promotes research on multimodal adversarial attacks, and helps enhance VLMS security for reliable real-world applications. The latest papers on VLMS adversarial attacks are continuously collected in https://github.com/chilljudaoren/A-Survey-of-Adversarial-Attack.

ACM Transactions on Privacy and Security
Georgia Institute of Technology (US), Chongqing University (CN), Zhejiang University of Science and Technology (CN), Shanghai Jiao Tong University (CN), Harbin Institute of Technology (CN), Zhejiang University (CN), Fuzhou University (CN), Xi'an Jiaotong University (CN), Northeastern University (CN)
Adversarial Robustness in Machine Learning
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon