BRA-Audit：基于累积暴露审计点放置的 LLM 多智能体系统预算运行时审计
原标题：BRA-Audit: Budgeted Runtime Auditing for LLM Multi-Agent Systems via Cumulative-Exposure Audit-Point Placement
AI 摘要
arXiv 论文提出 BRA-Audit，一种面向 LLM 多智能体系统的预算感知运行时审计框架。它将系统执行建模为动态依赖图，在固定审计预算下优化审计点放置，以最小化未检查暴露，并通过贪心调度优先审计高影响和长期未审计区域。实验表明，该方法在保持审计效果的同时，显著降低 token 消耗，接近无审计的干净基线性能。
正文节选
BRA-Audit: Budgeted Runtime Auditing for LLM Multi-Agent Systems via Cumulative-Exposure Audit-Point Placement Abstract LLM-based multi-agent systems (LLM-MAS) solve complex tasks through specialized collaboration, but inter-agent dependencies can propagate hallucinated or malicious outputs into system-level failures. Auditor agents mitigate these risks, yet existing strategies face an efficiency dilemma: end-only auditing reviews long trajectories and final outputs, potentially weakening audit