AI RESEARCH
Adversarial Attacks Against MLLMs via Progressive Resolution Processing and Adaptive Feature Alignment
arXiv CS.CV
•
ArXi:2605.09902v1 Announce Type: new Adversarial perturbations can mislead Multimodal Large Language Models (MLLMs) recognize a benign image as a specific target object, posing serious risks in safety-critical scenarios such as autonomous driving and medical diagnosis. This makes transfer-based targeted attacks crucial for understanding and improving black-box MLLM robustness.