Over the weekend, a massive privacy leak unfolded right in plain sight. Hundreds, possibly thousands, of shared conversations and interactive artifacts built within Anthropic’s Claude AI were indexed publicly on Google.
It started simply. A Reddit user typed site:claude.ai/share into Google.
The results were shocking.
Instead of seeing their own private notes, the user saw a scrollable list of public links. Some contained health records. Others had internal company documents. Several exposed the names and phone numbers of school-aged children.
This isn’t a glitch in the code. It’s a misunderstanding of how “sharing” works on the open web.
Зміст
The “Public” Trap
The core issue lies in a feature called Share Chat.
When you use this tool, you generate a URL. The interface warns you: “Anyone with the link can view.”
Most users interpret “anyone with the link” as meaning “my friends” or “my team.”
On the open web, “anyone” means literally everyone.
Google Docs, for instance, has similar sharing features. But it respects site-specific privacy controls. If you set a Google Doc to “private,” it doesn’t pop up in search results. It stays behind a login wall.
Claude’s share links are different. They are static pages on the web. Once indexed, they are public.
“We do not share chat directories or sitemaps… These shareable links are not guessable… unless people choose to share them.”
— Anthropic Spokeswoman Amie Rotherham
Anthropic’s defense is technical, not moral.
According to the company, they don’t hand search engines a list of private chats. The chats only appear in search if someone posts the link elsewhere publicly—like on a forum, Twitter, or a public dashboard. If you text a link to a friend, it stays private.
The problem? People are posting these links publicly. And Google indexes public things.
What Was Actually Exposed?
Before Google seemingly removed these pages, journalists and researchers dug in.
Futurism reported finding:
* Detailed medical reports of real patients.
* Clinical trial results including patient names.
* Documents containing contact info for primary school children.
* Internal employee reviews with personal data.
Fortune reported a chat labeled “shared by Anthropic” that generated erotica. This violates Anthropic’s usage policies against sexually explicit content. It’s unclear how the model bypassed its safety filters for this specific instance, or why the chat remained shared.
Anthropic has not commented on the specific case involving the explicit content.
Google’s Role
Ned Adriance, a Google spokesperson, pushed back on the idea that Google is solely to blame.
“Neither Google nor any other search engine controls what files are made public on the web,” Adriance told TechCrunch. “We give site owners clear controls… and we always respect those directives.”
Google argues that if a page is publicly accessible on the internet, it is fair game for indexing. If the site owner doesn’t use noindex tags, Google assumes it wants the page seen.
This isn’t the first time Claude has been indexed. Forbes reported a similar incident last year where Google indexed under 600 conversations. They disappeared quickly.
This time, the volume felt higher. Multiple users confirmed finding shared conversations using the same search operators.
How to Check Your Own Settings
If you’ve been using Claude, you might have accidental leaks.
To review which chats are set to public:
1. Go to Settings.
2. Navigate to Privacy.
3. Click on Shared Chats.
Here, you can review active links. If you see anything you don’t want on the open web, delete or disable it immediately.
The internet is louder than we think. Just because a feature is labeled “share” doesn’t mean it stays in the group chat. On the web, a share button is a broadcast tower.
Are you sure who’s listening?
The pages seem to have vanished as of Monday afternoon, per a TechCrunch test search. But the links may have already been scraped. Archived. Shared again.
The data is out there. It just isn’t visible to the casual browser anymore.
