arXiv Artificial Intelligence· Ashwin Nagarajan, Gursimran Panesar, Alicia Yang, Mona T. Diab, Virginia Smith, Aashiq Muhamed·· 2 小时前AI 评分40
Write Once, Check Many：面向智能体轨迹事后策略合规审计的可复用账本
Write Once, Check Many: Reusable Ledgers for Post-Hoc Policy Compliance Auditing of Agent Trajectories
AI 导读
研究提出“write once, check many”框架，由写入模型将每条智能体轨迹转为与策略无关的账本，读取模型仅凭账本核查新策略，无需重读原始轨迹。
来源：arXiv Artificial Intelligence · arxiv.org