Towards Effective MLLM Jailbreaking Through Balanced On-Topicness and OOD-Intensity

Open in new window