Your Shared Claude Chats Ended Up on Google. Audit What Your Team Pastes In.

Over one weekend, a Reddit user found a trick. Type site:claude.ai/share into Google and up popped real conversations people thought were private. Some had medical records. Some had internal company docs. This is not new, and it is not just Claude. Let me catch you up on what actually happened and what to check before your next review.

A share link that meant more than people thought

Claude has a feature that lets you share a chat. It makes a public page and hands you a link. The interface says "Anyone with the link can view." People read that as "anyone I send it to." It really meant anyone, full stop, including Google's crawler.

TechCrunch found the fix was simple and missing: a noindex tag would have told search engines to skip those pages. Google Docs uses the same kind of share flow and those files don't show up in search. The gap here was a default, not a hack. Anthropic told reporters that links only get indexed when someone posts them somewhere public, and that a link sent privately stays out of search.

What was actually in those chats

This is where it stops being abstract. Futurism reported finding a detailed medical report on a real patient, clinical trial results with patient names, documents listing the names and phone numbers of grade-school kids, and files marked internal use only.

Mashable noted two more: a user building a crypto wallet who exposed their keys, and a lawyer asking whether they had to self-report a breach of conduct. These are the exact things people paste into a chatbot because it feels like a private helper. It is not one. Every share turns a chat into a web page that lives outside your control.

This keeps happening, and not to one company

The pattern is old. Last year, Forbes reported a similar Claude exposure, with Google estimating it had indexed just under 600 conversations. ChatGPT had its own version last August, when OpenAI pulled the share feature after chats showed up in search. A researcher scraped around 100,000 public ChatGPT conversations the same way.

So this is the second time for Claude and at least the third public incident across major tools. And even chats you never shared don't fully leave. Lifehacker points out Anthropic holds your data for 30 days after you delete a conversation, and unless you opt out, your chats can be used to train future models. Some companies have humans review them.

The default is the risk

Step back and the border case makes the same point from the other side. The government is prosecuting an American for wiping his phone at an airport. Security expert Runa Sandvik's takeaway was blunt: it's better to not have that data on you when you cross certain borders. Download what you need once you arrive.

Same lesson for AI tools. The safest data is the data you never put in the box. Your team can't rely on a vendor's default to protect what shouldn't be there in the first place.

The deep cut

Anthropic said sharing works as intended, and Google said it respects whatever crawl rules a site sets. Both are technically right, and that is exactly the problem. The trust gap sits in the space between two correct answers, and your team is standing in it. Nobody catches that gap for you.

So do the boring thing this week. Open Settings, Privacy, Shared Chats in Claude and audit every public link your team made. Write one rule people can remember: no patient data, no customer PII, no keys, no unreleased docs pasted into any AI tool, shared or not. Assume anything shared can be indexed and can be scraped, because it has been, three times now.

Three questions for your team

1. Who on our team has created public share links in Claude or ChatGPT, and can we produce that list by Friday?
2. What are we pasting into these tools right now that would embarrass us or break a contract if it showed up in a Google search?
3. Have we turned off training data use and written a one-page rule for what never goes into an AI tool, and does everyone actually know it?