Vis-Poison: Poisoning Visual Knowledge in Multimodal Retrieval-Augmented Generation
This work introduces Vis-Poison, a novel visual knowledge poisoning attack where the poisoned image itself is the attacker-controlled payload, without manipulating captions, summaries, metadata, or other associated text.
Ru-Jin Liang, Zhongpu Chen, Yuhao Lei et al.
· 0 citations