---
id: 20260812-T0-01
title: "Anthropic、OpenAI和Google的思维链加密可被重放窃取，论文揭示漏洞"
title_en: "Researchers Steal Reasoning Traces from Anthropic, OpenAI, and Google APIs via Replay Attack"
url: https://ai.daily.yangsir.net/daily/20260812-T0-01
issue_date: 2026-08-12
publish_date: 2026-08-11T22:40:45.000Z
category: research
source_name: "Simon Willison"
source_url: https://simonwillison.net/2026/Aug/11/stealing-reasoning-traces/#atom-everything
---

# Anthropic、OpenAI和Google的思维链加密可被重放窃取，论文揭示漏洞

一篇新论文揭示：Anthropic、OpenAI和Google返回给客户的加密思维链（chain-of-thought）数据存在严重安全漏洞。攻击者可跨会话、跨用户甚至跨模型重放这些加密块，从而窃取前沿模型的推理过程。论文网站stolen-thoughts.com已上线，相关技术细节已公开。

## English Version

**Researchers Steal Reasoning Traces from Anthropic, OpenAI, and Google APIs via Replay Attack**

A new paper reveals that encrypted chain-of-thought blocks returned by Anthropic, OpenAI, and Google can be replayed across sessions, users, and models, allowing attackers to steal reasoning traces from frontier models. Details published at stolen-thoughts.com.

---

**来源**：[Simon Willison](https://simonwillison.net/2026/Aug/11/stealing-reasoning-traces/#atom-everything)

**详情页**：https://ai.daily.yangsir.net/daily/20260812-T0-01

---

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