analiza aiqbench
This research exposes critical vulnerabilities in frontier multimodal large language models by demonstrating how a single adversarial image can successfully fool diverse systems in black-box settings. The proposed O-Attack framework significantly increases attack success rates on leading models like GPT-5.4 and Claude-4.6, highlighting the urgent need for more robust semantic alignment and defense mechanisms against transferable perturbations.
One Attack to Fool Them All: Highly Transferable Black-Box Adversarial Attacks on Frontier MLLMs
arXiv:2609.33833v1 Announce Type: cross Abstract: Adversarial attacks have long posed a fundamental threat to machine learning systems. As multimodal large language models (MLLMs) rapidly evolve and become widely deployed, assessing their vulnerability to such attacks is essential for their safe use. In this work, we investigate whether a single adversarial image can consistently mislead diverse frontier MLLMs in black-box settings. We propose O-Attack, a highly transferable black-box attack framework. This framework builds on our insight that surrogate models contain a broad, high-level, cross-modally aligned semantic space. This space extends beyond final-layer outputs and provides multiple semantically consistent representations that remain underexploited by existing attacks. Within this space, O-Attack anchors aligned representations, progressively broadens semantic conditions, and optimizes perturbations through semantic consensus to promote consistent target alignment. By fully exploiting this space with the same surrogate models as M-Attack, O-Attack raises attack success rates on GPT-5.4 (29.1% to 77.2%), Claude-4.6 (42.8% to 81.6%), and Gemini-3.1 (38.2% to 80.9%). Extensive experiments across 24 MLLMs show that O-Attack outperforms six state-of-the-art methods in black-box transferability, with consistent effectiveness across prompts and improved efficiency and imperceptibility. This work exposes the practical safety risks posed by black-box adversarial attacks against frontier MLLMs, underscoring the need for more rigorous robustness evaluation and more effective defenses.
Abstract:Adversarial attacks have long posed a fundamental threat to machine learning systems. As multimodal large language models (MLLMs) rapidly evolve and become widely deployed, assessing their vulnerability to such attacks is essential for their safe use. In this work, we investigate whether a single adversarial image can consistently mislead diverse frontier MLLMs in black-box settings. We propose O-Attack, a highly transferable black-box attack framework. This framework builds on our insight that surrogate models contain a broad, high-level, cross-modally aligned semantic space. This space extends beyond final-layer outputs and provides multiple semantically consistent representations that remain underexploited by existing attacks. Within this space, O-Attack anchors aligned representations, progressively broadens semantic conditions, and optimizes perturbations through semantic consensus to promote consistent target alignment. By fully exploiting this space with the same surrogate models as M-Attack, O-Attack raises attack success rates on GPT-5.4 (29.1% to 77.2%), Claude-4.6 (42.8% to 81.6%), and Gemini-3.1 (38.2% to 80.9%). Extensive experiments across 24 MLLMs show that O-Attack outperforms six state-of-the-art methods in black-box transferability, with consistent effectiveness across prompts and improved efficiency and imperceptibility. This work exposes the practical safety risks posed by black-box adversarial attacks against frontier MLLMs, underscoring the need for more rigorous robustness evaluation and more effective defenses.
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2609.33833 [cs.CV]
(or arXiv:2609.33833v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.33833 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 304 cuvinte
Subiecte conexe
SemanticAdv: Generating Adversarial Examples via Attribute-conditional Image Editing
arXiv · 16 sept. 2026
Transferable Spatial Temporal Coherence Adversarial Attack on Black-Box Vision Language Models for Autonomous Driving
arXiv · 7 oct. 2026
How Vulnerable Is My Learned Policy? Universal Adversarial Perturbation Attacks On Modern Behavior Cloning Policies
arXiv · 6 oct. 2026
Adversarial Attacks on Online Handwriting using Salience-based Temporal Editing
arXiv · 22 sept. 2026