When LLM Meets DRL

In this paper, they propose RLbreaker, a black-box LLM jailbreaking attack driven by deep reinforcement learning (DRL).