Fetching the paper…

GradSafe: Detecting Jailbreak Prompts for LLMs via Safety-Critical Gradient Analysis · Around