GMAE: Gated Multi-signal Adaptive Ensemble for Targeted Transfer Attacks on Commercial Large Vision-Language Models
Open Access
- 15 June 2026
- conference paper
- Published by Association for Computing Machinery (ACM)
- p. 1974-1983
Abstract
Despite strong performance on open-source large vision-language models (LVLMs), transfer-based targeted adversarial attacks often fail against closed-source commercial LVLMs. Existing methods typically rely on static perturbation allocation and uniformly weighted surrogate ensembles, limiting their ability to capture model-agnostic semantic vulnerabilities essential for cross-model transferability. Consequently, perturbations are frequently allocated to semantically irrelevant regions, reducing effectiveness under realistic black-box settings. To address this limitation, we propose the Gated Multi-signal Adaptive Ensemble (GMAE) framework, a novel transfer-based targeted attack that formulates adversarial generation as a joint optimization over a gated perturbation field and an adaptive ensemble objective. A learnable spatial gating mask dynamically reallocates perturbation budgets toward semantically vulnerable regions, while surrogate models are adaptively reweighted according to optimization difficulty and historical consistency, encouraging alignment on shared semantic weaknesses across heterogeneous architectures. To further enhance robustness, GMAE performs global semantic alignment in a symbolic discrete space, reducing the influence of non-essential feature variations and improving cross-model generalization. In addition, we introduce a local attention-guided multi-scale consistency mechanism that transfers fine-grained target textures into attention-identified sensitive regions, ensuring both visual imperceptibility and semantic coherence at feature and concept levels. Extensive experiments demonstrate that GMAE achieves strong transferability to commercial LVLMs, and reasoning-oriented models. Notably, GMAE achieves an attack success rate of 82.90% on GPT-4o, outperforming the strongest baseline by 7.8%, establishing a new state of the art for targeted transfer attacks on large-scale black-box LVLMs.Keywords
Funding Information
- Basic scientific research business expenses in Xinjiang Uygur Autonomous Region (XJEDU2025P134)
This publication has 13 references indexed in Scilit:
- Jailbreaking Multimodal Large Language Models via Consistent Cross-Modal Backgrounds2025
- Adversarial Attention Perturbations for Large Object Detection Transformers2025
- Anyattack: Towards Large-scale Self-supervised Adversarial Attacks on Vision-language Models2025
- Efficient Generation of Targeted and Transferable Adversarial Examples for Vision-Language Models via Diffusion ModelsIEEE Transactions on Information Forensics and Security, 2024
- Enhancing Visual Question Answering through Question-Driven Image Captions as Prompts2024
- Improved Baselines with Visual Instruction Tuning2024
- A Survey of Multimodel Large Language Models2024
- ReActNet: Towards Precise Binary Neural Network with Generalized Activation Functions2020
- Towards Evaluating the Robustness of Neural Networks2017
- Microsoft COCO: Common Objects in Context2014