Fetching the paper…

FigStep: Jailbreaking Large Vision-Language Models via Typographic Visual Prompts · Around