According to the BBC, several hundred private conversations with the Claude AI were discoverable on the public web, and many of those links could be found via Google search.
The issue stemmed from user-created shareable links for conversations with Claude. These links are intended to let users share AI chats with others, but if a link was exposed publicly, search engines could index it. In such cases the conversations might have contained sensitive corporate or personal information if users had shared those details with the chatbot.
This is not an entirely new problem: OpenAI previously experienced a similar situation when Google indexed shareable links to AI conversations. Anthropic’s spokesperson told the BBC that Claude users ultimately decide whether to share their chats.
Over the weekend, Anthropic disabled the ability for conversations to be searchable, and the company says this should prevent private chats from appearing in Google results going forward.
Google commented that it does not police which pages become public on the internet, and that responsibility lies with site owners.
