Private conversations with the AI Claude published online

Hundreds of private conversations with Claude, Anthropic’s famous AI chatbot, were recently accessible to everyone online. These conversations, which sometimes contained personal and professional information, suddenly appeared in Google search results when users performed a specific search query.

The issue stemmed from the fact that users had the option to share a link to their conversation. Search engines like Google indexed these links, making them accessible to the general public. Anthropic has since resolved the issue by removing the links from search results, but many conversations had already been saved and shared.

A spokesperson for Anthropic emphasized that users themselves decide whether and when to share their conversations. “Links cannot be guessed or found unless users share them themselves,” she said.“When someone shares a conversation, that content becomes publicly accessible and, like any other public web content, can be archived by third parties.”

Be careful with the “share” feature

Claude’s share feature does warn that “anyone with the link” can view the content, but it does not explicitly mention that the link may appear in search results. Reddit users discovered these public discussions, which spanned more than 200 conversations spread across at least 25 pages of search results—some dating back only a few weeks. The conversations ranged from trivial questions to sensitive information. For example, users shared resumes containing names and contact information and conducted confidential searches, including transcripts of private conversations related to healthcare. One user even asked Claude how he could turn intoa “nine-legged fox.”

This incident is reminiscent of previous issues encountered by OpenAI’s ChatGPT m Elon Musk’s Grok, where chat histories were also made public. Google stated that it has no control over publicly accessible pages, but that website owners can decide for themselves whether their content is indexed.