Anthropic's Claude AI conversations appeared in Google searches

Thousands of conversations between Anthropic's AI chatbot Claude and its users, alongside tools created during those chats, were discoverable through Google searches earlier this month. The issue, likely involving shared and publicly set conversations, has since been rectified by Anthropic.

Among the conversations that appeared in Google search results were reportedly sensitive materials, including sexting and a project intended to train future security engineers. Embarrassing or potentially private conversations were shared across social media platforms over the weekend until Anthropic delisted the affected Google search results.

This incident echoes a similar security issue encountered by OpenAI's ChatGPT nearly a year ago. In July 2023, Fast Company revealed a comparable instance where shared ChatGPT conversations were indexed by search engines. OpenAI subsequently removed the ability to allow search engine indexing of conversations.

Cybersecurity analyst Rachel Tobac highlighted that users clearly do not understand the implications of the conversations they share. "It's an obvious user experience issue," Tobac stated. "Users clearly do not understand what they are agreeing to or that they are publishing their passwords, keys, personal questions, and proprietary data to the world."

Anthropic maintains that users have control over sharing their conversations publicly and that the company does not share chat directories or sitemaps with search engines. The company asserts that links are only discoverable when a user deliberately shares them, at which point the conversation becomes public and can be archived by third-party services.