Exploiting Inconsistencies to Bypass Multimodal Large Language Models

SI-Attack models 多模态 大型语言模型 安全机制 有害指令