Americans pouring their private thoughts, medical details, and financial data into AI chatbots just learned the hard way: nothing you type into Silicon Valley's machines stays between you and the screen. Over the weekend, users discovered that hundreds of Claude conversations — including health records, crypto wallet keys, children's names and phone numbers, internal company documents, and API credentials — were fully searchable on Google and Bing.
The exposure wasn't a hack. It was a feature working as designed — which is exactly the problem. Claude's "share chat" button creates a public URL with the warning that "anyone with the link can view." Users reasonably assumed that meant sending a link to a friend or coworker, not broadcasting it to the entire internet. But once those URLs landed anywhere a search engine crawler could reach — a forum, a social media post, a group chat — Google and Bing vacuumed them up like any other webpage.
Anthropic's response? Blame the user. Spokeswoman Amie Rotherham told TechCrunch that "when someone shares a conversation, they are making that content publicly accessible, and like other public web content, it may be archived by third-party services." Google's response? Blame Anthropic. Spokesperson Ned Adriance told both TechCrunch and WIRED that "neither Google nor any other search engine controls what pages are made public on the web" and that site owners have controls to prevent indexing.
Neither company wants to own the fact that Americans' most intimate data ended up on the open internet. The technical reality, as WIRED reported, is more damning than either admits: Anthropic relied on a robots.txt file to block crawlers — a request search engines can and do ignore. Google's own documentation states it will bypass robots.txt if a page is linked from elsewhere and lacks a "noindex" HTML tag on the individual page. Anthropic didn't include that tag on shared chat pages. WIRED confirmed this by reviewing the exposed pages. This is not a mystery. It's a choice.
It's also a repeat. Forbes reported the same problem with Claude in September 2025. The same thing happened with OpenAI's ChatGPT last year, when roughly 100,000 conversations became searchable. A researcher scraped those exposed chats, meaning the data persists regardless of whether Google removes the results. As 404 Media noted, third parties may have already grabbed the Claude data too.
The scope of what was exposed is staggering. According to reporting across all four outlets, indexed conversations included detailed medical reports with patient names, clinical trial results, documents with children's contact information, employee reviews with personal details, resumes, business documents marked internal, political party advice, legal ethics questions, erotica — including at least one chat labeled "shared by Anthropic" that appeared to show Claude generating explicit content in violation of its own policy, per Fortune — and an AI therapy app someone had built.
As of Monday afternoon, the Google search operator surfacing these chats no longer returns results, suggesting Anthropic or Google quietly remediated the issue. Bing still showed roughly 612 results at the time of WIRED's reporting.
The real question isn't why this happened again. It's why anyone in Silicon Valley expects Americans to trust them with their data when the business model depends on hoovering up every conversation, every thought, every query — and the safeguards are a suggestion, not a guarantee.