Anthropic's Claude private chats appear publicly, prompting security scrutiny

What This is

• Some Claude chat pages and user-made “artifacts” became searchable on Google and other search engines after people used Claude’s share feature.
• The indexed pages included conversations that appeared to contain sensitive information like personal details, work documents, and even crypto wallet keys.
• The exposure was tied to public share URLs, not to private chats inside a user’s Claude account, and Anthropic said it was not a system breach.
• Search visibility happened in part because the shared pages did not include “noindex” tags, which tell search engines not to list a page in results.
• Anthropic had relied on robots.txt to discourage crawling, but that approach does not guarantee pages will not be indexed if other sites link to them.
• Anthropic said shared links are public by design and not “guessable,” and that it does not submit directories or sitemaps to search engines.
• Anthropic moved to block indexing and many results were removed from search listings, but previously shared links may still work if someone has the URL or an archive exists.
• Guides recommend users review Claude’s privacy settings to find and revoke shared chats and shared artifacts, and avoid sharing anything they would not want public.

Why This Matters

• It highlights how a “share” button can turn private-seeming AI use into public data in minutes.
• It shows that small web settings can have major privacy effects when search engines and archives copy public pages.
• It raises trust questions for workplaces that paste proprietary or regulated information into AI chats.
• It adds pressure on AI firms to use safer defaults and clearer warnings when making content public.
• It reminds users that “public link” often means “publicly discoverable,” not just “anyone with the URL.”.