Anthropic’s AI chatbot Claude has come under scrutiny after some user conversations appeared in Google and Bing search results, raising concerns about AI privacy. However, the incident was not a hack or a data breach. Instead, it involved conversations that users had intentionally shared using Claude’s public Share feature, making them accessible through a public web link.
The issue has renewed debate over how AI companies handle shared conversations and user privacy.
Were Private Chats Really Exposed?
The answer is partly yes, but not in the way many people assumed.
Only chats that users had deliberately shared using Claude’s Share option were affected. This feature creates a public webpage containing a snapshot of the conversation, allowing others to view it through a unique link.
The problem arose when some of these public pages became indexed by Google and Bing, making them discoverable through search instead of only by people who already had the link.
What Information Was Found?
Researchers reported finding publicly searchable conversations containing:
- API keys
- Cryptocurrency wallet information
- Résumés and CVs
- Company documents
- Legal discussions
- Personal addresses and phone numbers
- General personal conversations
While these chats had been intentionally shared at some point, many users were unaware that search engines could potentially index the links, making the information easier to discover.
Why Did Search Engines Index Them?
Anthropic had instructed web crawlers not to crawl shared chat pages using its robots.txt file.
However, security researchers noted that the pages reportedly did not include a “noindex” tag, which is one of the standard methods search engines use to prevent webpages from appearing in search results.
As a result, if a shared Claude link appeared elsewhere on the internet—such as on social media or a forum—it could still be indexed by search engines.
Following public attention, Google removed the indexed results, although existing shared links remained accessible to anyone who already had the URL.
Anthropic’s Response
Anthropic clarified that:
- Claude conversations remain private by default.
- Only chats users intentionally shared through the Share feature were affected.
- The company does not submit shared chat directories or sitemaps to search engines.
- Users can review or revoke shared conversations at any time through Settings → Privacy → Shared Chats.
The company has also taken steps to reduce the chances of shared chat pages appearing in search results again.
A Reminder About AI Privacy
The incident highlights an important lesson for AI users.
Sharing an AI conversation creates a publicly accessible webpage, and anything included in that conversation—whether personal details, confidential business information or access credentials—may become visible to others if the link is exposed or indexed.
Users should avoid including sensitive information in conversations they intend to share publicly and regularly review any previously shared AI chats.
Summary
Anthropic’s Claude did not suffer a traditional data breach, but some conversations shared through its public Share feature appeared in Google and Bing search results. The affected chats included sensitive information such as API keys, company documents and personal details. Anthropic says private chats remain private by default, and users can delete or revoke shared conversation links through their privacy settings.
