Anthropic says it fixed a privacy exposure in which some Claude chatbot conversations accessed via the “share” feature appeared in public Google search results. Reporting traced the issue to snapshots that were meant to be shared with recipients via link, but search engines could surface the shared conversation pages when users used a specific Google query pattern. The exposed chats reportedly included sensitive information such as cryptocurrency wallet keys and personal details (names and addresses), plus content that included explicit material that conflicts with Anthropic policy. Anthropic said its privacy principles do not include sharing chat directories or sitemaps with Google and characterized the links as non-guessable except by users choosing to share. The episode is a direct compliance and student-safety signal for higher education institutions adopting generative AI tools, including policies on when learners and staff can use “share” features, how logs are handled, and what training is needed to prevent sensitive data from being published inadvertently.