Mohan Kankanhalli

Overloading Large Vision-Language Models for Jailbreaking

Haoyu Zhang, Yangyang Guo, Mohan Kankanhalli

Large Vision-Language Models (LVLMs) exhibit remarkable vision-language capabilities and are increasingly deployed in real-world applications such as personal assistants, document analysis systems, and embodied agents. However, their dual-modal attack surfaces make them vulnerabl…Read more