Jailbreaks on Vision Language Model via Multimodal Reasoning

Aarush Noheria, Yuguang Yao

Vision-language models (VLMs) have become central to tasks such as visual question answering, image captioning, and text-to-image generation. However, their outputs are highly sens…