SAFESHIELD：小语言模型部署时安全的决策组织框架
原标题：SAFESHIELD: A Decision-Organization Framework for Deployment-Time Safety of Small Language Models
AI 摘要
该论文提出 SAFESHIELD，一个面向小语言模型部署时安全的决策组织框架。作者将部署时安全形式化为决策组织问题，包含面向责任的决策分解与显式协调两个维度，并实例化了准入、路由、证据、发布四类决策职责，通过可审计的 Decision Trace 记录决策。实验表明，移除准入门控会显著增加错误发布，而阻断上游证据会使发布准确率从 96.0% 降至 69.5%，说明部署时安全不仅取决于单个护栏能力，也取决于决策如何被组织与协调。
正文节选
SAFESHIELD: A Decision-Organization Framework for Deployment-Time Safety of Small Language Models Abstract Deployment-time safety of language models is commonly implemented through runtime guardrails such as input moderation, routing, retrieval verification, and output filtering. Existing deployment frameworks provide increasingly capable mechanisms for these functions, but offer limited guidance on how the safety decisions they produce should be explicitly organized, coordinated, and audited. W