This paper shifts the object of role engineering from human users to large language model (LLM) agents, and proposes PRISM to address the over-authorization …
机构:北大
来源:arXiv 2611.11539 | AI4Papers 论文推荐平台