---
id: 20261002-T0-11
title: "AI智能体可借共享缓存互相传递指令，形成蠕虫式攻击链"
title_en: "AI Agents Can Pass Instructions via Shared Cache, Forming Worm-Like Attack Chains"
url: https://ai.daily.yangsir.net/daily/20261002-T0-11
issue_date: 2026-10-02
publish_date: 2026-10-01T06:29:01.000Z
category: news
source_name: "Simon Willison"
source_url: https://simonwillison.net/2026/Oct/1/matthew-green/
---

# AI智能体可借共享缓存互相传递指令，形成蠕虫式攻击链

密码学家Matthew Green引用研究指出，AI智能体攻击已形成完整链路：一个载荷劫持智能体，被劫持的智能体再将载荷传递给下一个智能体。实验发现，运行在独立隔离沙箱中的智能体可以通过共享的软件包缓存互相留下指令，从而实现跨沙箱传播。这意味着即使每个智能体单独隔离，共享基础设施仍可能成为攻击载体。对开发者而言，多智能体系统的安全设计需要重新审视沙箱隔离的有效性。

## English Version

**AI Agents Can Pass Instructions via Shared Cache, Forming Worm-Like Attack Chains**

Cryptographer Matthew Green highlights research showing a complete AI agent attack chain: a payload hijacks an agent, and that agent carries the payload to the next one. Experiments found that agents running in separately isolated sandboxes could leave instructions for each other in a shared package cache, enabling cross-sandbox propagation. This means that even with individual agent isolation, shared infrastructure can still serve as an attack vector. For developers, the security design of multi-agent systems needs to reconsider the effectiveness of sandbox isolation.

---

**来源**：[Simon Willison](https://simonwillison.net/2026/Oct/1/matthew-green/)

**详情页**：https://ai.daily.yangsir.net/daily/20261002-T0-11

---

*智语观潮 · Daily — https://ai.daily.yangsir.net/llms.txt*