这是 AI Agent 论文播报 在 2026-05-29 的论文播报。本期内容由 AI 自动生成,欢迎留言交流。
本期重点
- How Coding Agents Fail Their Users: A Large-Scale Analysis of Developer-Agent Misalignment in 20,574 Real-World Sessions
- WorldMemArena: Evaluating Multimodal Agent Memory Through Action-World Interaction
- The Best-Laid SCHEMEs: Coordinated Sabotage and Monitoring in Multi-Agent Systems
['今天主线:安全防合谋、记忆做阶段诊断、真实日志驱动评测']
