In early August 2026, Anthropic confirmed that some Claude conversations people had shared using the app's built-in "Share" feature were showing up in Google search results. Security researcher Bruce Schneier's blog, citing reporting on the incident, described what turned up: a vibe-coded AI therapy app, meeting notes, a medical billing dashboard, private cryptocurrency wallet keys, and people's home addresses — all readable by anyone who happened to search the right terms, because the shared link had been indexed like any other public webpage.

Anthropic's position, as reported, is that shared links "are not guessable or discoverable unless people choose to share them themselves" — meaning the company doesn't consider this a bug in the traditional sense. Once you create a share link and it becomes reachable from the open web, it can get crawled, indexed, and even archived by third parties, the same as any other public page. That's a reasonable technical description of how the internet works. It's also not what most people think they're agreeing to when they tap a button labeled "Share" inside what feels like a private chat window.

This isn't unique to Claude, either. If you've used ChatGPT's own conversation-sharing feature at any point, it's worth applying the same scrutiny — sharing tools across AI chat products tend to work the same way: generate a public link, and "public" means public to search engines too, not just to the one person you meant to send it to. The fix isn't to stop using share features. It's to know exactly what you've shared, check whether any of it is now sitting in a search index, and get in the habit of treating the share button as "publish," not "send."

Why this happens

When you share a conversation, the app doesn't email a private copy to your recipient. It generates a URL pointing to a real, hosted page containing your full conversation, and that page lives on the open internet. Whether that page ends up in Google depends on things outside your control: whether the site's indexing rules changed, whether someone linked to your share URL from elsewhere, whether a browser extension or crawler picked it up, or whether an archiving service snapshotted it before anyone caught the problem. You can behave perfectly and still end up indexed if the platform's defaults shift under you — which is roughly what happened here.

The practical implication: don't assume a shared link is safe just because you didn't post it anywhere public yourself. Assume it's public from the moment you create it, and audit accordingly.

Step 1: Find every conversation you've ever shared

Most people have no idea how many share links they've generated, because the action is easy to forget — you click "Share" once to send a friend a chatbot answer, and never think about it again.

  • Claude: Look in your account or privacy settings for a list of shared conversations or shared links (menu wording changes over time — search the settings panel for "share" if you don't see it immediately). If there's no central list, the fallback is to open each conversation you remember sharing and check its sharing status directly from within that chat.
  • ChatGPT: Check Settings → Data Controls (or the equivalent "shared links" area — OpenAI has moved this menu before). It should show every conversation you've ever generated a public link for, even ones you shared months or years ago and forgot about.
  • Any other AI tool you use for anything sensitive (coding assistants, note-taking AI, customer-support bots with a "share transcript" option) — repeat the same check. If a tool has a share button, it has this same class of risk.

Go through the list and ask, for each one: would I be comfortable with a stranger reading this if they found it through a search engine? If the answer is no, move to Step 3 now — don't wait until you've finished checking Google first.

Step 2: Check whether you're already indexed

Before you assume nothing has leaked, search for yourself the way a stranger would.

  • In Google, search a distinctive phrase from a conversation you shared — something specific enough that it wouldn't appear on any other page, like an unusual project name, a rare error message, or a full sentence you typed. Put it in quotes to force an exact match.
  • Search your own name or email address alongside the platform's domain, e.g. "your full name" claude.ai or "your full name" chatgpt.com.
  • If you've shared anything containing an account number, a physical address, or a wallet key, search for that string directly (carefully — don't paste live credentials into a search box on a shared computer). If it turns up, you need to rotate or invalidate that credential regardless of what happens to the search result.

This won't catch everything — Google doesn't index the entire web instantly, and some exposure lives in archives (like the Wayback Machine) rather than live search. But it's the fastest way to find out if something's already out.

Step 3: Revoke the link and request removal

Two separate things need to happen, and doing only the first one isn't enough:

  1. Turn off sharing at the source. Inside the conversation (or the shared-links list from Step 1), disable or delete the share link. This should make the URL stop working going forward — but it does not automatically erase any cached or indexed copy that search engines or archivers already grabbed.
  2. Ask Google to drop the already-indexed page. If a search result for your content is still showing up after you've revoked the link, Google's "Remove outdated content" tool (accessible via a search for that name, or through Google Search Console if you manage a verified property) lets anyone request removal of a URL that no longer resolves or that contains content you've deleted. This speeds up de-indexing rather than waiting for Google's crawler to notice on its own.

Keep in mind that revoking and de-indexing can't undo a page that a third party has already archived or copied elsewhere — which is exactly the risk Anthropic pointed to when it said shared content "may be archived by third parties like other public web content." Treat anything you've ever shared as potentially permanent, even after you delete it.

Step 4: Decide what never gets shared again

Going forward, the simplest rule is the safest one: if a conversation contains anything you wouldn't want a stranger, an employer, or a future search of your own name to surface, don't use the share button on it — full stop. That includes:

  • Anything resembling therapy, medical, or mental-health conversations, even ones that started as casual venting.
  • Financial details, crypto wallet keys or seed phrases, account numbers, or tax information.
  • Your home address, a family member's name, or anything else that identifies where you live.
  • Draft business plans, internal meeting notes, or anything with a client's or employer's confidential information in it.

If you want to show someone a chatbot response without the public-link risk, take a screenshot or copy the relevant text into a message instead. It's slightly less convenient than clicking "Share," and that's the point — it keeps the decision to publish in your hands instead of a platform's indexing settings.

The checklist

  1. Find your shared-links list on every AI chat tool you use, not just one.
  2. Search your own distinctive phrases and name against each platform's domain to check for existing exposure.
  3. Revoke any share link you don't actively need public.
  4. Request removal of anything already indexed, and treat it as potentially archived regardless.
  5. Rotate any credential, key, or password that ever appeared in a shared conversation — do this regardless of whether it shows up in a search.
  6. From now on, share screenshots or copied text for anything sensitive, and reserve the live share-link feature for conversations you'd be fine with anyone reading.