Primary sourcerohanpaul_ai
View sourceEvent ·Multi-source
Salesforce 论文:智能体互相协作时从不主动核实对方说法
Salesforce 发布新论文,研究代表不同利益方的智能体之间的协作失败问题。团队用 Claude Code、Codex 和 Cursor 在隔离沙箱里跑了超过 2,500 次智能体对智能体的会话,覆盖 488 个场景,归纳出 20 种失败模式。结果显示没有任何智能体会重新拉取文档去核实同伴的事实性声明,主动联系同伴基本只在提示词要求时才发生。在一个服务纠纷案例中,一笔 24,000 美元的和解通过了全部 10 项结果检查,同时供应商方的智能体泄露了机密信息,暴露出团队整体会继承最弱智能体的错误。
Current read
Salesforce 跑了 2500 多场智能体互相对话的实验,发现它们互相不查证还泄密,做多智能体系统的都该看看。
11 sources57° AI heatLast updated 2026/10/19 17:04:38
Evidence
11 sourcesRelated source 1Ate-a-Pi
View sourceRelated source 2掘金本周最热
View sourceRelated source 3kimmonismus
View sourceRelated source 4techcrunch
View sourceRelated source 5Anyscale
View sourceRelated source 6Latent.Space
View sourceRelated source 7Andrew Curran
View sourceRelated source 8The Information
View sourceRelated source 9Viking
View sourceRelated source 10Notion
View sourceConflict check
Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.