Anthropic’s Claude links indexed on Google, exposing chats

On July 25 a Reddit user discovered that hundreds of share links from Anthropic’s Claude chatbot had been indexed by Google because the pages lacked a noindex meta tag. The indexed pages included private conversations, apparent Social Security numbers, crypto wallet seed phrases, legal consultations and internal documents.

Claude’s share feature creates URLs labeled “Anyone with the link” that are intended to be unlisted. Anthropic had a robots.txt file blocking Claude’s share URLs, but robots.txt stops crawlers from fetching pages and does not prevent a search engine from indexing a URL if the link appears on other sites. In some search results Google showed the links with the message “No information is available for this page” when it could not access the page contents.

Searches for pages under claude.ai/share and claude.ai/public/artifacts returned payroll spreadsheets with employee names, customer relationship exports, unreleased product roadmaps and a range of personal chats. A Reddit thread that flagged the issue drew thousands of upvotes and examples of sensitive material posted to the shared links.

Third parties archived many of the exposed conversations. A public repository titled Shared-Claude-Chats collected 453 Claude conversations and 519 Grok chats, totaling 11,241 messages in plain text. Copies stored by archives or other sites can remain available after the original links are removed.

Anthropic updated its configuration and added noindex tags on July 26. Google began removing the indexed results after the change; Bing continued to surface shared links at the time of reporting. Anthropic had not released a public statement by publication.

Users who want to stop future access to a shared conversation can revoke the share in the Claude app under Settings → Privacy → Shared Chats. Revoking a link prevents further access via that URL but does not delete copies already archived or indexed by other parties.

Robots.txt blocks crawlers from fetching pages but does not stop a URL from being listed in search results when the link appears elsewhere on the web. A similar indexing issue affected another AI chat product about 11 months earlier when a discoverability option caused large numbers of conversations to become searchable.