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

Yueqi Xie, Minghong Fang, Renjie Pi, Neil Gong

Browse the full ACL paper archive.