Anthropic Blames Users After Claude Chats Leak Medical Records, Kids' Info

A fresh wave of exposed Claude conversations surfaced over the weekend, and the scale is still an open question. Reddit users discovered that Google queries like “site:claude.ai/share” pulled up everything from clinical trial results with patient names to the phone numbers of elementary school children. Futurism confirmed finding a detailed real patient medical report and internal-only company documents in the search results. It’s a near-identical repeat of a 2023 incident where Google indexed roughly 600 Claude chats before they vanished. This time, Anthropic’s defense lands squarely on users.

Spokeswoman Amie Rotherham told TechCrunch that shared links aren’t guessable and stay out of search engines “unless people choose to share them themselves.” The logic is straightforward: Claude’s share feature works like Google Docs, and if someone drops the link on a public forum, search crawlers find it. Anthropic says it doesn’t submit chat directories or sitemaps to Google. The company draws a hard line — when you hit share, you’re publishing to the web, and third-party archiving is just what the web does. Google’s spokesperson echoed the point, noting site owners have clear controls to block indexing.

But there’s a gap between what the interface implies and how people actually behave. “Anyone with the link can view” is a warning, sure — but it’s the same one Google Docs displays, and nobody expects their private documents to pop up in search results. The mental model for most users is that an unlisted link is obscure, not truly public. Last year, a researcher scraped about 100,000 public ChatGPT conversations, proving this isn’t just an Anthropic problem. It’s an industry-wide assumption that users understand the permanence of a URL, and the evidence keeps suggesting they don’t.

As of Monday afternoon, the Google search trick stopped returning results, so the immediate leak appears plugged. But one embarrassing artifact lingers: Fortune spotted a chat labelled “shared by Anthropic” in which Claude generated erotica — content explicitly banned by the company’s usage policy. Anthropic hasn’t commented on that specific case, and how the content was prompted remains unclear. You can check your own exposure by heading to Settings > Privacy > Shared Chats. The fix might be simple, but the underlying tension between sharing features and user expectations isn’t going anywhere.