Anthropic's Claude shared chats surfaced in Google search
TL;DR
- Reddit users found that a Google 'site:claude.ai/share' query returned more than 200 shared Claude conversations across at least 25 pages of results.
- The indexed pages reportedly contained health records, unpublished corporate documents, CVs, and in some cases the names and phone numbers of children.
- Anthropic says shared links are not guessable and only appear public if users choose to share them; the listings were removed over the weekend.
A share button that most people probably read as 'send this to a friend' turned out to mean 'publish to the open web,' and by the weekend Reddit users were pulling up strangers' Claude conversations through a single Google search operator. The BBC reported that more than 200 shared chats surfaced across at least 25 pages of Google results before the listings were removed.
What people had unknowingly published is the part that stings. According to the reporting, the indexed pages included health records, unpublished corporate documents, CVs with contact details, and in some cases the names and phone numbers of children. TechCrunch's writeup adds clinical trial documents with patient names and internal employee reviews to that list. One user, apparently not expecting anyone else to read the exchange, asked Claude 'do you want to help me or do you want to help anthropic more?'
Anthropic's line, given to both outlets by a spokeswoman, is that this is user-driven: shared links are 'not guessable or discoverable unless people choose to share them themselves,' and once shared, 'they are making that content publicly accessible, and like other public web content, it may be archived by third-party services.' Google's response was the parallel one, that search engines do not decide what a site publishes. Both are technically correct. Both also sidestep the UX question, which is whether a link labelled 'anyone with the link can view' actually reads to a normal user as 'this may be crawled by Google.'
This is not the first time it has happened, and that is what should make product leaders uncomfortable. Forbes reported last year that hundreds of Claude chats had already been indexed and scrubbed. OpenAI hit a nearly identical issue with ChatGPT and tightened its controls, and Grok reportedly had hundreds of thousands of chat logs made publicly searchable in 2025. The honest caveat is that the reporting does not tell us how many of the exposed users were on enterprise plans, whether Anthropic has notified anyone, or what specifically the company is going to change in the share flow.
The follow-through worth watching is who moves first: whether Anthropic uses the moment to ship a genuinely private-by-default share, with expiring links, noindex on by default, and one-click revoke, or whether a rival gets to tell that story instead.
Shared on Bluesky by 2 AI experts
-
"The share option within Claude tells a user that 'anyone with the link' may view the contents of that link, but does not explicitly state that the link may end up in Google and search results." Internet scholars have b…
View on Bluesky →
Originally reported by bbc.co.uk
Read the original article →Original headline: Some people's chats with Claude AI found publicly available online