Google Indexed Hundreds of Claude Share Links. Every AI Tool You Use Has the Same Button.

A Reddit user ran a single site query against Claude's share subdirectory and found the results populated with other people's chats and Artifacts. Screenshots spread across Reddit and X within hours.

Reporting from TechCrunch, VentureBeat and Neowin described what was in them: résumés, financial spreadsheets, internal business discussions, and legal questions. Several outlets went further, reporting credentials and health information among the exposed pages — those specific claims come from smaller publications and secondhand screenshots, so treat the worst of it as unverified. The verified core is bad enough.

This part is worth understanding even if you never touch Claude, because the same trap sits in a lot of small SaaS products.

Claude's share pages carried two directives that quietly cancelled each other out:

Think about what you actually put into an AI chat in a given week. Your pricing model and the margins behind it. Churn numbers you haven't told anyone. Draft positioning for a launch that hasn't happened. A customer email you're trying to word carefully. Cofounder equity math. A contract you're nervous about.

Now think about how often you hit Share. Solo founders share more than employees at bigger companies, not less. There's no internal wiki, no Confluence page, no team channel — so the chat link becomes the deliverable. You send it to a contractor, an advisor, a Discord, a client. Each one of those is a publish action against a URL you don't control and will never think about again.

That's the real exposure here, and it isn't specific to Claude. Every AI tool in your stack has a share button, and each one made its own decision about what that button means.

Here's the part most of this week's coverage skipped, and it's the reason you shouldn't file this under "resolved."

We ran this exact play eleven months ago. In August 2025, Fast Company found more than 4,500 shared ChatGPT links in Google, some containing names, résumés and confidential work content. The mechanism was different — a "Make this chat discoverable" checkbox in the share dialog, with the consequence explained in smaller grey text underneath. OpenAI's CISO Dane Stuckey said they'd pull the chats from search engines, and the company killed the feature, calling it a short-lived experiment.

Then this happened: over 100,000 of those conversations had already been archived on the Internet Archive. They are still there. OpenAI has not asked for their removal. De-indexing solved the discovery problem and did nothing about the existence problem.

Do this once, properly, and then make it a habit. It genuinely takes ten minutes.

Share links aren't going anywhere. They're one of the best organic growth loops an AI company has — every shared conversation is a product demo with the recipient's trust already attached. No vendor is removing that button.

What should change is the defaults. Expect expiring links, workspace-scoped sharing, per-share visibility controls, and noindex-by-default headers to become table stakes over the next few product cycles. OpenAI removing the discoverability toggle entirely was the first move in that direction; this week was the second.

The structural issue is slower and more uncomfortable. Your AI chat history has quietly become a business record — it holds more unfiltered detail about your company than your CRM does — and almost nobody has a retention policy for it. If you handle customer data, that's not just a privacy question, it's a compliance one. The founders who get ahead of this will be the ones who decided what belongs in a chat window before the next incident, not after.