Users have discovered that some public chats with Claude, Anthropic's artificial intelligence, have started appearing in Google search results, raising concerns about privacy. It was expected that even when using public links, a certain level of privacy would be maintained.

This incident has once again drawn attention to how AI platforms communicate the degree of privacy and security for the content they generate. The issue is related to Claude's sharing feature, which allows users to create public URLs for chats and artifacts, including documents, spreadsheets, web apps, and other AI-generated content. By default, chats remain private, but when a public link is created, the content becomes accessible to anyone. As a result, some of these pages have been indexed by search engines, including Google, and are now available through search queries.

Anthropic stated that it does not provide search engines with directories or sitemaps containing user chats, and only those chats intentionally shared via public links can become publicly accessible. The company notes that by default, Claude chats remain private unless a user creates a public link. Meanwhile, Google reminded that website operators themselves determine whether public pages will be indexed by search engines. Claude's public pages did not include 'noindex' metadata, which is typically used to prevent indexing, allowing some chats to appear in search results.

Among the indexed pages, there could be materials containing sensitive information, such as internal documents, legal discussions, and personal conversations. There is no evidence that Claude disclosed private chats without user action, but this case demonstrates how easily public sharing features can be mistakenly perceived as tools for restricted access. Similar concerns have previously arisen with other AI platforms that allow sharing chats via public links.

Users are advised to treat a link to a shared conversation as they would any other public web page, unless explicit security measures are in place. This incident also highlights the need for AI platforms to clearly distinguish between "general access" and "public searchability," especially as generative AI tools become more integrated into professional and collaborative workflows.