面向可信系统的生成式AI：健康检查模型
原标题：Generative AI for trustworthy systems - Towards a health check model
AI 摘要
该论文基于对电信、汽车、国防、航空、银行、能源、政府和软件服务等八个行业十八位资深从业者的访谈，提出「可信自主性健康检查模型」。该模型用八个维度（系统层的智能体权限范围、保障机制、数据可信度、架构遏制、可追溯性与可理解性；组织层的治理、人类监督姿态、劳动力能力可持续性）和四种信任范式（运营、工程、统计、遏制）取代传统单维成熟度模型。作者强调模型是诊断性而非规定性的，并形式化提出「对齐假设」：有效可信度受限于智能体权限范围与支撑维度之间的匹配程度。
正文节选
∎ Generative AI for trustworthy systems Abstract The adoption of generative AI in software-intensive systems is proceeding rapidly across diverse industrial contexts, but the analytical instruments currently used to characterize that adoption — principally unidimensional maturity models — compress important configurational variation into a single progressive axis. Drawing on an inductive interview study of eighteen senior practitioners across telecommunications, automotive, defence, aviation, ba