Claude Chats on Google: Sensitive Data Exposed by Users

Last week, sensitive personal information from Claude chats appeared on Google. This included medical reports, clinical trial results, and internal company files. Here's the thing: these documents were shared by users themselves, not due to a hack. Users clicked a "share" button in Claude, which makes the conversation a public web page. While users might intend to share a link with a colleague, search engines like Google then index these pages. Redditors first alerted the public, and Google began removing pages around July 26th. This isn't an isolated incident; it's the fourth time a chatbot has exposed sensitive user data in a year. OpenAI and Grok previously had similar issues. What's interesting is that companies like Anthropic state they don't hand Google directories of shared chats, and links aren't guessable. However, if a user posts a shared link where a search crawler can find it, the content becomes public. The bottom line: users are inadvertently exposing private information by using these "share" buttons, highlighting a recurring problem with how public sharing is understood and implemented in AI tools.