Hundreds of people's private Claude conversations were found sitting in Google search results

[Tel92]

Hundreds of user conversations with Anthropic's Claude chatbot were found to be publicly searchable through Google and other search engines, exposing personal and professional information that users had believed was shared only with specific recipients. Reddit users first discovered the issue over the weekend, finding more than 200 conversations spread across at least 25 pages of search results, some from chats that had taken place just weeks earlier. The conversations included people seeking help with their CVs, complete with names, contact details and work history, and in some cases what appeared to be proprietary research for their jobs, including healthcare related material

The root cause was Claude's share chat feature, which lets a user generate a link intended only for people they specifically send it to. Despite that intent, many of these links were automatically indexed by search engines and left accessible to anyone who searched for them, without the original user ever realizing it had happened. Anthropic removed the chats from search indexes over the weekend, though many had already been saved and shared elsewhere before that happened. A Google spokesperson told the BBC the company does not control what pages are made public on the web, saying that responsibility lies with individual site owners, and that Google gives site owners clear controls to decide whether pages get crawled or indexed

This is not a new problem for the AI industry. OpenAI experienced an almost identical issue last year when nearly 100,000 ChatGPT conversations became publicly accessible through search, after which the company changed how easily such logs could be indexed. Grok, the chatbot built into Elon Musk's X platform, saw an even larger scale version of the same problem last year, with hundreds of thousands of chat logs made publicly searchable. The recurring pattern across three separate major AI chatbots suggests share features designed with a specific recipient in mind keep getting undermined by default search engine indexing behavior that neither users nor, in some cases, the companies themselves fully anticipated

Some people's chats with Claude AI found publicly available onlinewww.bbc.com

[Oscar_57]

The pattern repeating across Claude, ChatGPT and Grok in successive years shows this isn't a one-off engineering oversight, it's a genuinely recurring blind spot in how share link features get designed across the whole industry