Guard models are increasingly used to safeguard LLM-based agents, primarily by identifying actions that agents are forbidden to perform. However, identifying…
机构:清华
来源:arXiv 2610.11773 | AI4Papers 论文推荐平台