Thousands of private conversations held with Claude are currently findable and readable online, with some containing sensitive personal details including medical information, legal questions, and financial data. The issue, first reported by Futurism, stems from Claude's shared chat feature, which allows users to generate public links to their conversations. Many of those links have since been indexed by search engines, making the chats discoverable to anyone who knows how to look.

The exposure is not a traditional data breach in the hacking sense. Users who clicked "share" on their chats generated publicly accessible URLs, and search engine crawlers did the rest. Still, it is clear that many people did not understand the implications of sharing those links, and the content now surfacing online suggests they had no intention of making their conversations public. This story follows earlier reporting on Claude shared chats appearing in Google Search results, which first flagged the indexing problem.

What's Actually Exposed

The range of content found in these publicly accessible chats is broad. Some conversations include users describing personal health symptoms and asking for medical guidance. Others contain details about ongoing legal situations, employment disputes, and relationship problems. A number of chats show users sharing what appear to be proprietary business strategies or internal company details. In short, these are the kinds of conversations people typically have in confidence, and the assumption of privacy appears to have been reasonable given how the sharing feature was presented.

Key Facts

  • Claude's shared chat feature generates public URLs that search engines can index.
  • Exposed conversations include medical, legal, financial, and personal content.
  • Users appear to have shared links without understanding the public indexing risk.
  • The issue is distinct from a data breach; it results from user-generated public links.
  • Anthropic has been under pressure to clarify its data handling practices following the report.

Privacy advocates have pointed out that the burden of understanding indexing behavior should not fall entirely on users. Most people interacting with an AI assistant are not thinking about web crawlers. The default expectation when sharing a link with a specific person is that only that person sees it, not that it becomes a permanent, searchable public record. Anthropic responded to the situation after the search indexing issue gained wider attention, though questions remain about what steps will be taken to protect users going forward.

"People are putting incredibly sensitive stuff into these AI chats and they have no idea it could end up on Google."Futurism, quoting a privacy researcher familiar with the issue
Claude AI Handboek by Leon Tindemans
Get the Claude AI Handboek
458 pages on getting more out of Claude, by AI expert Leon Tindemans. A printed book, written in Dutch, shipped worldwide with track and trace.
View the book →

Anthropic's Position and the Path Forward

Anthropic has maintained that users who generated these links made them public themselves. That is technically accurate, but critics argue the company has a responsibility to make the consequences of sharing far more explicit, and to consider whether search engine indexing should be blocked by default for shared chats. Other platforms that offer link-sharing, including some Google products and Notion, use noindex tags or other mechanisms to prevent shared content from appearing in search results unless the user actively opts in to full public visibility.

This incident adds to a broader conversation about how AI companies handle user data. Anthropic has previously accused Alibaba of harvesting millions of Claude conversations, which underlines just how valuable these chat logs are and why protecting them matters. The company is navigating a difficult balance between building collaborative and shareable features and ensuring users genuinely understand the risks involved.

For users who have previously shared Claude chat links, the practical advice is to revisit those links and disable sharing if the conversations contain anything sensitive. There is currently no indication that Anthropic will retroactively block indexing of already-shared chats, though that may change as pressure mounts. Anyone using Claude should treat the share button with the same caution they would apply to posting on a public website, because for now, the outcome is effectively the same.

Further reading: Learn more about Claude's model family, read our background on Anthropic, or browse the latest Claude AI news.