Overview
- Users and researchers found dozens of Claude conversations indexed by Google and Bing after a simple site search exposed pages created by the chatbot’s “Share” feature.
- The indexing happened because Anthropic relied on a site-level robots.txt request instead of a server-side noindex header so pages could be listed if their share URL appeared elsewhere on the open web.
- The discoverable pages included high-risk material such as named patient data, internal corporate records, staff reviews, and live API keys which created real privacy and security harms.
- Search queries that returned Claude chats over the weekend of July 26–27 stopped showing results after search engines and Anthropic moved to de-index and add noindex headers, but any direct share link posted or copied earlier still opens the full chat.
- Regulators and organisations are treating the issue as an operational governance problem and recommend revoking old share links, adding server-side revocation and archival controls, training staff, and treating every share link as potentially public.