Claude chats exposed on Google are making some user conversations and creations discoverable through ordinary search, according to 404 Media, including material people may have meant to share with a small group rather than the open web.
The visible material reportedly includes an AI therapy app someone appears to have built with Claude, meeting notes, and a dashboard apparently made to inspect medical billing data. 404 Media also reported that some exposed chats contained sensitive information, including cryptocurrency wallet keys and personal details such as addresses.
Anthropic, the company behind Claude, did not immediately respond to 404 Media's request for comment.
Why are Claude chats exposed on Google?
Claude lets users create share links for conversations, a feature meant for sending a transcript to another person or a team. If a shared page is reachable on the public web and not blocked from indexing, search engines can list it, which means the audience can expand far beyond the people who received the link.
404 Media said a specific Google search technique, often called a Google dork, had surfaced Claude conversations. A Google dork is a tightly targeted search query that uses search operators or distinctive page patterns to find pages, files, or services that are not obvious through normal browsing.
A post on the Claude subreddit described one such search and said the results included chats with API keys, login credentials, names, addresses, and phone numbers. 404 Media reported that the search no longer returned results when it tested the query on Google and DuckDuckGo.
Claude's exposure is not limited to chat transcripts, according to 404 Media. The company also offers Artifacts, interactive workspaces that users can use to build or test small apps and tools. 404 Media reported that a separate search for these shared Artifacts was still producing results at the time of publication.
The risk is not just that a search result exists today. 404 Media compared the situation to a previous exposure involving OpenAI's ChatGPT, in which nearly 100,000 shared conversations were searchable on Google. In that case, a researcher was able to scrape exposed chats, which means removing pages from search results does not guarantee the material has disappeared from other people's copies.
Forbes previously reported that Claude transcripts had appeared in Google search results as well, according to 404 Media. The recurring pattern is blunt: sharing features that feel like private forwarding can behave like publishing when the resulting page is open to the web.
How can Claude users hide shared chats?
Anthropic's support documentation says Claude users can adjust sharing and privacy settings to make shared chats no longer publicly accessible. Users who have shared sensitive material should review their shared links, revoke access where possible, and treat exposed secrets such as API keys or wallet credentials as compromised.
This story draws on original reporting from 404 Media.
