EAIR：企业级 LLM 多智能体系统的交互风险框架
原标题：EAIR: Interaction Risks in Enterprise LLM Multi-Agent Systems
AI 摘要
来自微软、慕尼黑大学等机构的研究者提出 EAIR 框架，用于分析企业级 LLM 多智能体系统中的交互风险。该框架将企业交互风险形式化为工作流必须保持的组织属性，并给出可测试的风险分类、失败标准与证据要求。团队构建了 EAIR-Worlds 可执行微世界，并开发诊断与修复流水线，通过执行轨迹定位导致约束违反的交互边界并指导修复。大规模实验显示，局部合理的智能体行为会反复违反组织级约束。
正文节选
EAIR: Interaction Risks in Enterprise LLM Multi-Agent Systems Yuchen Ma Affiliation: LMU Munich Swadheen Shukla Affiliation: Microsoft Paulo Salem Affiliation: Microsoft Yue Huang Affiliation: University of Notre Dame Robert Sim Affiliation: Microsoft Stefan Feuerriegel Affiliation: LMU Munich Michel Galley Affiliation: Microsoft Jianfeng Gao Affiliation: Microsoft Abstract LLM agents increasingly divide enterprise work across roles with different goals, authority, access, and information. Becau