Claude AI leak: Hundreds of private chats with health, legal data indexed by Google and Bing

Discovery

Over the weekend, Reddit users stumbled on a privacy breach: typing the search operator “site:claude.ai/share” into Google displayed a long list of Claude conversations and Artifacts, interactive mini-apps and documents, that anyone could open. By Monday, the news was reported by 404 Media, Axios, WIRED, and other outlets, confirming that shareable links intended for limited circulation had been swept into the indexes of Google, Bing, Brave Search, Ecosia, and DuckDuckGo. Claude’s share button warns that “anyone with the link can view,” but does not explain that the link may end up in search engine caches.

We give people control over sharing their Claude conversations publicly, and in keeping with our privacy principles, we do not share chat directories or sitemaps with search engines like Google.

Why the links were indexed

Anthropic’s robots.txt file has told crawlers to stay away from shared chat URLs since at least September 2025, according to snapshots on the Wayback Machine. Nevertheless, both Google and Bing recommend that web developers also include a “noindex” meta tag on individual pages they want to keep out of results. WIRED examined a sample of the exposed Claude pages and found that none carried that tag. Google’s developer guide says it may ignore robots.txt if the page is linked from elsewhere and lacks an explicit noindex directive. Because people shared their Claude URLs on Reddit, X, and forums, those links were picked up by crawlers and passed into search results.

Neither Google nor any other search engine controls what pages become public on the web and these pages were indexed by many search engines.

Bing’s documentation mirrors this: blocking a crawler with robots.txt alone may not prevent indexing. As of Monday evening, a Bing search for “site:claude.ai/share” still returned about 612 results.

Sensitive data on display

The BBC tallied over 200 conversations across at least 25 pages of search results, but La Vanguardia put the total in the thousands. The content was startling. Futurism found a detailed patient medical report, clinical trial results with names, documents with primary-school children’s phone numbers, internal-only corporate files, and employee evaluations containing personal information. 404 Media uncovered a vibe-coded AI therapy app, meeting notes, and medical billing data. One artifact opposed the dissolution of a Miami city agency overseeing park management; another proposed changes to a university academic calendar. Business plans, multi-cloud infrastructure diagrams, and clinical-trial planning documents also surfaced. Users sought career advice and shared resumes with full contact details, asked about ethical obligations for attorneys in Kansas, role-played erotic scenarios, and debated political affiliations. A chat from April showed a user requesting to “become a nine-tailed fox,” with Claude claiming to give “fully functional fox powers.”

Clean-up and fallout

After the issue went viral, Anthropic moved to fix the exposure. By Monday afternoon, a TechCrunch test query on Google returned no results. The BBC noted that many chat logs had already been saved and shared widely online before they were pulled. Anthropic’s statement maintained that links are “not guessable or discoverable unless people choose to share them themselves.” The incident follows a similar lapse in June 2025, when Meta AI chatbot conversations accidentally appeared in a public feed. The Claude episode demonstrates that a feature designed for quick sharing can become a privacy hazard when the mechanics of web indexing are overlooked.