Pages created through the share feature in Anthropic's Claude were reachable from Google search results over the weekend of July 25 to 27, 2026. A specific search syntax was enough to open other people's conversations and Artifacts one after another. Some of them contained patient records and the contact details of primary school children, exposing a wide gap between what the share feature does and what users assumed it did.

It started with a weekend post on Reddit

The first person to notice was a Reddit user who posted on Saturday. Typing site:claude.ai/share into Google returned a long list of Claude share pages. By Monday morning, US outlets had picked up the story and attention spread quickly.

Claude lets users turn a conversation or a project into a shareable URL. The interface warns that anyone with the link can view it. Read plainly, that wording suggests a feature meant for handing a chat to a colleague or a friend. Google Docs offers a comparable link-sharing option, and documents shared that way do not end up in Google search. This time, links created with that same mental model were reachable by anyone through a search engine.

Text conversations were not the only thing affected. Artifacts, the small apps and documents users build inside Claude, were equally open, and work notes and code were reported among them.

The contents went well beyond casual chat

What made the situation serious was the nature of what turned up. A detailed medical report on a real patient. Clinical trial results with patient names attached. Documents listing the names and phone numbers of primary school-aged children. Company files marked internal use only. Employee reviews containing personal information about staff. None of it is material a passing stranger should be reading.

There were also reports of a chat carrying an Anthropic-shared label that contained sexually explicit output. Anthropic's usage policy explicitly bars Claude from generating that kind of material, and how the conversation came about was still unclear at the time of writing.

The scale of the exposure has not been independently confirmed. The closest reference point is a similar incident last year, when just under 600 conversations were estimated to have been indexed by Google. That same year, a researcher was reported to have collected roughly 100,000 publicly shared ChatGPT conversations. Whatever the exact numbers, the pattern of share links being harvested at scale is not new.

The two explanations do not quite meet

Anthropic's position is that the feature worked as designed. A company spokesperson said share links appear in search results only when the URL has been posted somewhere search engines can reach, such as a forum or a social media post, and that a link sent privately to one person stays out of search. The company added that it does not hand chat directories or sitemaps to search engines, that the links are not guessable, and that choosing to share a conversation makes it publicly accessible web content that third-party services may archive.

Google took a similar line. A spokesperson said search engines do not decide which pages are made public on the web, and that these pages had been indexed across many search engines. Site owners are given controls over crawling and indexing, and those directives are respected.

Both explanations hold up technically. The trouble is that the only sentence a user reads is the one saying anyone with the link can view the page. Getting from there to a conclusion about search engine indexing is a leap most people will not make. This was less a vulnerability than a mismatch between how a feature is worded and how far it actually reaches.

The search results are gone, but a check is still worth it

By Monday afternoon, following the steps described in the Reddit post no longer returned matching results, so the exposure itself appears to have been resolved. The share links themselves have not disappeared, however. Anyone holding a URL can still open that conversation.

Users can review which chats they have made public from within Claude. Opening the shared chats management screen under the privacy section of settings lists every conversation with an active link, and unwanted ones can be revoked on the spot. Anyone who remembers sharing work material or personal information should take a look.

Conversations with an AI assistant tend to carry more detail than a search history or a notes app, and the fact that the other party is not a person lowers the bar for what gets typed in. Before pressing share, it is worth asking whether the action is showing something to a handful of colleagues or publishing a page on the web.

Summary

Pages created by Claude share links were reachable through Google search on July 25 to 27, 2026, including conversations containing medical records and children's contact details. Anthropic says the feature behaved as designed, and Google says site owners decide what is public. The exposure appears to have been resolved, but the share links remain live. Open the shared chats list under the privacy section of settings and clear out anything that no longer needs to be public.

Source: [1] TechCrunch, "PSA: Your Claude shared chats and Artifacts may have ended up on Google"