Leave My Images Alone: Preventing Multi-Modal Large Language Models from Analyzing Images via Visual Prompt Injection
别让我看图片:通过视觉提示注入防止多模态大语言模型分析图片
机构 * Georgia Institute of Technology(佐治亚理工学院) ; Duke University(杜克大学)
专题命中 提示注入 :prompt injection(title,abstract);safety(abstract);分类 cs.AI、cs.LG
AI总结 本文提出ImageProtector,通过在图像中嵌入精心设计的微小扰动,使多模态大语言模型在分析时产生拒绝响应,同时评估了三种潜在的防御措施,发现它们在降低ImageProtector效果的同时影响模型性能。
Comments Appeared in ACL 2026 main conference
Journal ref The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026)