智能体团队为何跨边界失败？A2A系统私有状态失败分类
原标题：Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States
AI 摘要
该研究针对跨组织边界的智能体间（A2A）系统，分析了超过2500次运行记录，覆盖8个场景族（法律、CRM、安全、商务等），提出了包含20种失败模式的分类体系 -Failures，分为发现与委托、基础与记录一致性、验证、行为与忠实度四类。研究发现智能体倾向于匹配提示中提到的同伴名称而非主动搜索可用同伴、不会主动互相事实核查，且仅评估最终输出会掩盖内部协调与验证问题。作者据此呼吁对智能体进行个体化评估，并公开了运行日志数据集。
正文节选
Why Do Agent Teams Fail Across Boundaries? A Failure Taxonomy for Agent-to-Agent Systems with Private States Abstract Agents are increasingly interacting with other agents that act for different principals—people or organizations on whose behalf they work—e.g., a buyer’s personal agent negotiating with an enterprise’s seller agent. In such agent-to-agent (A2A) scenarios, each agent has its own environment and holds information and authority that the other cannot see. Moreover, the interests of e