CourionAI
EN
Newsletter
← All news
privacy 2 min read

Shared Claude chats turned up in Google, and the lesson applies to every chatbot

Conversations shared with Claude's link feature were briefly indexed by search engines because the pages lacked a noindex tag. Anthropic fixed it quickly. The bigger point: a share link is a public web page, whatever the app calls it.

Risograph illustration of a speech bubble pinned like a paper notice to a public bulletin board, a magnifying glass hovering above it

If you have ever pressed “share” on an AI conversation, this one is worth two minutes. Over the weekend, users found that shared Claude chats were showing up in ordinary Google results. A single search query, site:claude.ai/share, surfaced thousands of conversations that their authors almost certainly thought only their recipients would see. Anthropic appears to have fixed the problem quickly, and the Google results faded soon after. On Bing and Brave Search, the pages lingered longer.

The cause was small and unglamorous. Pages created by the “Share with link” feature were missing a noindex tag. That tag is a one-line instruction to search engines that says: you may fetch this page, but do not list it in your results. Without it, a crawler treats a share link like any other public web page and files it away. Nothing was hacked, no account was broken into, and no private chat was exposed. Only conversations that someone had deliberately turned into a link were affected. Reported contents included crypto keys and legal questions, which tells you how quickly “just sharing this with a colleague” turns into something you would rather not have archived.

This is the same mistake OpenAI made last year with shared ChatGPT chats, and it comes from a genuine ambiguity in the product. A share link feels private because you decide who receives it. Technically it is a public page with an address that is hard to guess, and “hard to guess” only protects you until something starts guessing, or crawling, or scraping. Security people call this security through obscurity, and it works right up until the moment it does not. Every AI assistant now ships a share feature, they all work this way, and the fix is a configuration detail that is easy to miss when you launch fast.

What this means for you: If you have ever shared a Claude conversation, open Settings, then Privacy, then Shared Chats, and delete any link you no longer need. Do the same housekeeping in whatever other assistant you use. Going forward, treat a share link as a small public web page rather than a private message. Before you create one, skim the conversation for the things that quietly accumulate in chats: API keys, passwords, client names, health details, contract terms. If any of that is in there, copy the text into an email instead. For most people this incident changes nothing about whether AI tools are safe to use. It changes one habit, which is a fair trade.

Sources

Source: https://the-decoder.com/shared-claude-chats-were-reportedly-showing-up-in-search-engines/

Next story

Cursor rebuilt SQLite with a swarm of agents, and the cheap models did most of it

Given only the 835-page manual, Cursor's new agent swarm wrote a working SQLite clone in Rust. Splitting the job between an expensive planner and cheap workers cut the bill from $10,565 to $1,339 at the same quality.

Risograph illustration of one large geometric node branching downward into a wide tree of many small identical nodes