Continuous Embedding Attacks via Clipped Inputs in Jailbreaking Large Language Models

Open in new window