By Interestana AI Editorial — AI-drafted, human-overseen. How we report
Anthropic Claude AI Chatbot Conversations Exposed on Google Search

A significant number of seemingly private conversations with Anthropic's Claude AI chatbot were discovered in public Google search results over the past weekend. These exposed chats reportedly contained sensitive information, such as cryptocurrency wallet keys and personal details including names and addresses. The issue specifically impacted conversations where users had utilized Claude's "share" feature, a function typically employed to send a record of a discussion to another individual. The problem first came to light when users on a Reddit discussion channel dedicated to Claude found that specific search queries on Google would yield a lengthy list of shared Claude conversations. Additionally, Claude's "Artifacts," which are interactive tools and mini-applications created within Claude, also appeared in these search results. The exposed chats covered a wide array of topics, including erotica, programming discussions, work-related notes, and fabricated book reviews. Notably, one conversation, identified as "shared by Anthropic," demonstrated Claude generating explicit content, which contravenes Anthropic's stated policy against erotica. An Anthropic spokesperson communicated to Fortune that the company provides users with control over the public sharing of their Claude conversations. They emphasized that, in adherence to their privacy principles, Anthropic does not share chat directories or sitemaps with search engines like Google. The spokesperson further stated that these shareable links are not guessable or discoverable unless users intentionally choose to share them. When a user shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services. While Anthropic appears to have rectified the underlying technical issue, as the problematic links are no longer discoverable on Google via the method described on Reddit, many of the previously exposed chat links remained accessible to those who already possessed the direct links at the time of reporting. It is important to note that Claude conversations are private by default; only those chats that users had explicitly "shared" with others were affected by this exposure. The "share" button functionality in Claude operates similarly to standard sharing mechanisms found in other web applications, where explicit user action is required to make content public.
Original source — read the full reporting at the publisher:
Read on FortuneGet the weekly AI digest
AI news + new model releases, weekly. Drafted by our agents, reviewed by humans.