Unbridled icarus: A survey of the potential perils of image inputs in multimodal large language model security

Y Fan, Y Cao, Z Zhao, Z Liu, S Li - 2024 IEEE International …, 2024 - ieeexplore.ieee.org
Multimodal Large Language Models (MLLMs) demonstrate remarkable capabilities that
increasingly influence various aspects of our daily lives, constantly defining the new …

Jailbreakzoo: Survey, landscapes, and horizons in jailbreaking large language and vision-language models

H **, L Hu, X Li, P Zhang, C Chen, J Zhuang… - arxiv preprint arxiv …, 2024 - arxiv.org
The rapid evolution of artificial intelligence (AI) through developments in Large Language
Models (LLMs) and Vision-Language Models (VLMs) has brought significant advancements …

Instructta: Instruction-tuned targeted attack for large vision-language models

X Wang, Z Ji, P Ma, Z Li, S Wang - arxiv preprint arxiv:2312.01886, 2023 - arxiv.org
Large vision-language models (LVLMs) have demonstrated their incredible capability in
image understanding and response generation. However, this rich visual interaction also …

Test-time backdoor attacks on multimodal large language models

D Lu, T Pang, C Du, Q Liu, X Yang, M Lin - arxiv preprint arxiv …, 2024 - arxiv.org
Backdoor attacks are commonly executed by contaminating training data, such that a trigger
can activate predetermined harmful effects during the test phase. In this work, we present …

Sa-attack: Improving adversarial transferability of vision-language pre-training models via self-augmentation

B He, X Jia, S Liang, T Lou, Y Liu, X Cao - arxiv preprint arxiv:2312.04913, 2023 - arxiv.org
Current Visual-Language Pre-training (VLP) models are vulnerable to adversarial examples.
These adversarial examples present substantial security risks to VLP models, as they can …

Ot-attack: Enhancing adversarial transferability of vision-language models via optimal transport optimization

D Han, X Jia, Y Bai, J Gu, Y Liu, X Cao - arxiv preprint arxiv:2312.04403, 2023 - arxiv.org
Vision-language pre-training (VLP) models demonstrate impressive abilities in processing
both images and text. However, they are vulnerable to multi-modal adversarial examples …

Agent smith: A single image can jailbreak one million multimodal llm agents exponentially fast

X Gu, X Zheng, T Pang, C Du, Q Liu, Y Wang… - arxiv preprint arxiv …, 2024 - arxiv.org
A multimodal large language model (MLLM) agent can receive instructions, capture images,
retrieve histories from memory, and decide which tools to use. Nonetheless, red-teaming …

Probing the robustness of vision-language pretrained models: A multimodal adversarial attack approach

J Guan, T Ding, L Cao, L Pan, C Wang… - arxiv preprint arxiv …, 2024 - arxiv.org
Vision-language pretraining (VLP) with transformers has demonstrated exceptional
performance across numerous multimodal tasks. However, the adversarial robustness of …

Multimodal large model pretraining, adaptation and efficiency optimization

L Ji, S **ao, J Feng, W Gao, H Zhang - Neurocomputing, 2025 - Elsevier
Multimodal large models, leveraging extensive datasets and parameters, have provided
superior solutions for multimodal tasks and have been widely applied across various …

VADS: Visuo-Adaptive DualStrike attack on visual question answer

B Zhang, J Li, Y Shi, Y Han, Q Hu - Computer Vision and Image …, 2024 - Elsevier
Abstract Visual Question Answering (VQA) is a fundamental task in computer vision and
natural language process fields. The adversarial vulnerability of VQA models is crucial for …