PSA: Your Claude shared chats and Artifacts may have ended up on Google
TechCrunch Lorenzo Franceschi-Bicchierai ● Covered by 2 sources
Claude share links got indexed by Google, exposing health records and kids' info. Anthropic says it's not their fault—users made them public.
Based on reporting by TechCrunch, Lorenzo Franceschi-Bicchierai — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Over the weekend, Reddit users figured out that typing search operators like "site:claude.ai/share" into Google pulled up a long list of Claude conversations and Artifacts — the interactive mini apps and documents Claude lets people build — that were never meant for public eyes. Some of what turned up included health records, internal company documents, and the names and phone numbers of children.
The root of the problem traces back to Claude's "share chat" feature, which generates a link that lets anyone who has it view a conversation or project. The interface itself warns "Anyone with the link can view," phrasing that reads like it's meant for passing a chat to a friend or coworker, not broadcasting it to the entire internet. Google Docs has a similar sharing option, and those documents don't typically end up crawlable on Google — which is exactly the comparison that makes this look like a design problem rather than a one-off mistake.
Anthropic's explanation put the responsibility back on users. The company told TechCrunch that share links only show up in search results if they've been posted somewhere search engines can already see, like a forum or social media post, and that a link sent privately stays out of search entirely. Spokeswoman Amie Rotherham added that Anthropic doesn't hand chat directories or sitemaps over to search engines, that the links aren't guessable, and that once someone shares a conversation, it becomes public content like any other web page — meaning it can get archived by outside services.
What actually got exposed is uncomfortable reading. Futurism reported finding a detailed medical report tied to a real patient, clinical trial results listing patient names, documents naming primary-school-aged children along with their phone numbers, internal-use-only company files, and employee reviews containing personal information. Some exposed Artifacts held code and private work notes. Fortune also found a chat labeled "shared by Anthropic" in which Claude generated erotica — content explicitly barred under Anthropic's own usage policy — though it's not clear how that particular output was produced, and Anthropic hadn't responded to questions about it as of Monday.
A Reddit user flagged the issue Saturday, 404 Media reported it Monday morning, and by Monday afternoon TechCrunch's own attempt to replicate the original search operator returned nothing, suggesting someone quietly cleaned things up. Google's Ned Adriance noted that the company doesn't control what page owners choose to make public, and that it respects whatever crawling and indexing directives site owners set.
This isn't the first time a chatbot's sharing feature has leaked into search results. Forbes reported a similar situation last year involving Claude, when Google had indexed just under 600 conversations before they vanished from results — though it's unclear whether this new exposure matches that scale, even if the same search trick surfaced both. And separately last year, 404 Media reported that a researcher scraped around 100,000 ChatGPT conversations that users had set to share publicly. Anyone who has used Claude's share feature can check what's exposed under Settings, then Privacy, then Shared Chats.
My take — AI-written commentary, not fact-checked reporting
Blaming users for a sharing button that behaves like it's semi-private until it suddenly isn't is a neat trick, but it doesn't hold up — the whole point of a share feature is that people will actually use it to share things. Anthropic's defense that links only appear in search if posted elsewhere skips over the fact that forums and group chats are exactly where people paste links when a product tells them "anyone with the link can view." This is the second time in roughly a year that a major AI company's sharing tool has spilled private conversations into public search results, and until these features default to genuinely private rather than technically-private-but-easily-public, calling it a user error is just corporate spin.
Read more about this at: TechCrunch
