Event ·Multi-source

Anthropic 新研究：多智能体系统中的“思想病毒”传播机制

Anthropic 在 arXiv 2608.10218 发表论文，研究可自我传播的“思想病毒”如何通过多智能体系统扩散。研究让一支小规模智能体团队共事一个编码项目，再由上下文被清空的智能体链接力，发现共享工作产物可携带并传递恶意指令。实验显示传播效果受宿主模型、现有指令、载荷危害性和网络拓扑影响，有害载荷传播较差但仍会偶尔成功。若在系统提示中加入简短警告，几乎可以完全免疫。

Current read

Anthropic 做了个实验：让“思想病毒”在多个 AI 智能体之间传播，一句系统提示警告就能几乎完全免疫，论文在 arXiv 上。

11 sources67° AI heatLast updated 2026/8/12 16:20:02

Evidence

11 sources

Primary sourceelvis

2026/8/12 16:20:02View source

Related source 1Anthropic: Research

2026/8/13 10:00:39View source

Related source 2arXiv cs.AI

2026/8/10 17:24:50View source

Related source 3Gary Marcus

2026/8/13 13:47:31View source

Related source 4techcrunch

2026/8/13 18:28:14View source

Related source 5IT之家

2026/8/10 23:56:58View source

Related source 6小互

2026/8/13 14:15:19View source

Related source 7Decoder

2026/8/14 16:06:32View source

Related source 8宝玉

2026/8/10 19:48:03View source

Related source 9量子位

2026/8/11 03:51:24View source

Related source 10Aman Sanger

2026/8/11 17:31:22View source

Conflict 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.

Open the story