Con Instruction: Universal Jailbreaking of Multimodal Large Language Models via Non-Textual Modalities

Open in new window