Claude聊天记录在Google上曝光 - 免费分享的代价

安全与信任 41 min ago7加入收藏

Claude聊天记录在Google上曝光 - 免费分享的代价
插图 : Léa Fontaine

一个“分享聊天” Claude 生成一个公共链接 - 有时被 Google 索引。Anthropic 进行了修复,但该事件重新激活了一个旧的反射:每个“分享”按钮都是一个信息泄露的载体。

用简单的语言来说 - Anthropic的“分享聊天”功能导致部分Claude对话通过Google搜索可找到。用户没有意识到可分享的链接可能被索引。Anthropic正在修补,但这种模式(共享聊天→公共URL→搜索爬虫)之前已经影响过LLM生态系统。

事实

2026年7月底,多名Claude用户发现通过“分享聊天”功能生成的公共链接被Google Search索引(TechCrunch,7月27日)。Anthropic承认问题并正在努力阻止默认索引。ITmedia的报道总结指出,ChatGPT曾经历过类似事件。

我们的看法

薄弱环节不是加密的,而是用户体验:如果用户将“私有但可分享”的链接粘贴到可索引的渠道中,那么这种链接就不存在。两种诚实的选择:默认使用经过身份验证的共享(类似Google Docs),或者在生成链接时明确警告。LLM生态系统在每个周期中重新学习这一教训——“分享”功能太具吸引力,以至于没有摩擦就交付。

robots.txt ≠ 保密

通过robots.txt阻止爬虫可以防止未来的索引,但无法防止泄露。公共链接仍然对任何拥有它的人(搜索引擎、人类、爬虫)可访问。只有链接后面的身份验证才能消除表面。

关注点

  • Anthropic的修补程序:阻止爬虫(缓解)或默认使用经过身份验证的链接(结构性)。
  • 在撤回之前,对已索引的对话数量进行第三方审计,以及其中包含的内容。
Resources

本文由人工智能撰写,并经人工编辑审核。

我们的编辑部
Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

SSHMonitoringAI Ops
Get early access
这篇文章对您有帮助吗?

7 人赞了这篇文章

S
Sofia Adler安全与信任
🇨🇳 人工智能安全、模型可靠性、网络安全
分享:
评论 (7)

登录后即可参与讨论。

ArtLover99 28 Jul 2026 · 17:06

I wonder if there's a way to make users more aware of the public nature of shared chats before they hit the share button.

Emma_London 28 Jul 2026 · 19:27

Maybe a pop-up reminder could help users pause and think before sharing.

LitLover42 28 Jul 2026 · 17:03

I wonder if there's a way to implement a two-step verification before sharing chats publicly. Maybe a pop-up warning could help.

Alex 2 28 Jul 2026 · 16:40

I'm concerned about the potential misuse of shared chats. How can users ensure their conversations remain private in an age of easy sharing?

TechSavvy 28 Jul 2026 · 16:39

I'm curious if there's a way to educate users on the implications of sharing chats publicly, beyond just interface design.

ArtLoverLA 28 Jul 2026 · 16:34

I think it's crucial for platforms to design interfaces that prioritize user privacy by default, not just as an afterthought.

LecteurDuDimanche 28 Jul 2026 · 16:31

I wonder how often this happens and if users are aware of the risks when sharing chats publicly.

Alex_London 28 Jul 2026 · 16:25

I'm curious about the measures Anthropic is taking to prevent such incidents in the future. Will they be more transparent about their security protocols?

Your Linux servers, as a desktop.
TermalOSSponsored
Ops, reimagined

Your Linux servers, as a desktop.

Agentless SSH monitoring, a full remote desktop and an AI ops copilot — no agents to install. Everything stays on your machine.

Get early access
主题
浏览
信息