arXiv · Multiagent Systems· Shaswata Mitra, Raj Patel, Subash Neupane, Sudip Mittal, Md Rayhanur Rahman, Shahram Rahimi·· 3 小时前AI 评分42
多智能体系统安全的判断边界如何测量：规则与评判的分工
Where Rules End and Judges Begin: Measuring the Judgment Boundary in Multi-Agent Systems Security
AI 导读
研究提出 DEFER1（DEterministic-First Enforcement with Residual judgment）框架，将多智能体系统防御组织为五项原则，用 28 项级联检查拦截可判定攻击，其余交由四个评判模型处理。
来源：arXiv · Multiagent Systems · arxiv.org