Claude privacy just became a real problem. Over the weekend, anyone who typed "site:claude.ai/share" into Google could pull up thousands of actual Claude conversations, including legal strategies, health records, and private company documents. Not exactly what you’d expect from a chat that felt private.
What actually happened
Claude has a "share" feature. It generates a public link to any conversation so you can send it to a coworker or friend. The problem is those links were never truly private. They were regular web pages that search engines could crawl.
A Reddit user discovered this over the weekend and posted about it. Almost immediately, people found conversations that contained sensitive information. Lawyers discussing case strategies. Engineers working through proprietary technical problems. Even personal details like names and phone numbers of children.
The core Claude privacy issue is not new. This same thing happened back in September 2025. Anthropic patched it, or so everyone thought. Now it has happened again. Twice in under a year is not a coincidence. It is a pattern.
How Claude chats ended up on Google
Here is the technical part, kept simple. Anthropic used a file called robots.txt to tell Google and Bing not to index the /share/ directory. Think of robots.txt as a "do not enter" sign on a website. It works, kind of, except when it does not.
The problem is that Google explicitly states it will ignore robots.txt if a page is linked from somewhere else on the internet AND the page lacks a "noindex" tag. Bing says the same thing. Anthropic had the "do not enter" sign up. But they never put locks on the individual doors.
Anthropic’s Claude privacy setup relied on robots.txt alone. The individual shared chat pages did not have noindex meta tags. So when someone posted a Claude share link on a forum, Twitter, or Slack, Google followed that link, saw no noindex instruction, and added the page to search results.
This is not an obscure technicality. It is web development 101. Any developer who has dealt with SEO knows you need both robots.txt and noindex. Wired confirmed that a sample of exposed Claude chat pages had no noindex tag. The Claude privacy failure was entirely preventable.
Anthropic’s response missed the point
When asked about the Claude privacy leak, Anthropic’s spokesperson said that shareable links "are not guessable or discoverable unless people choose to share them themselves." In other words, they blamed users.
But that is not the full story. At least one affected user during the September 2025 incident said they never used the share feature. Their conversation still showed up in search results. Whether that is true this time around is unclear, but Anthropic’s blanket "it is your fault" response ignores the possibility that something else is going on.
Also, Anthropic said they do not "share chat directories or sitemaps with search engines." Sure. But they also did not properly protect the pages themselves. The Claude privacy setup was like putting a "keep out" sign in front of your house but leaving every window wide open.
As of this writing, Google no longer shows results for the search. Anthropic moved fast to get the pages deindexed after the story blew up. Bing, however, still showed around 612 results according to Wired’s report.
Is your data safe?
Probably, but probably is not reassuring. Here is the honest breakdown of Claude privacy risk right now.
If you never used the share feature, you were likely not affected. The leaked conversations came from public share links. But given that at least one user in the previous incident claimed they never shared anything, it is worth being cautious.
Even beyond this leak, there is a bigger Claude privacy concern that most people ignore. Everything you type into Claude is stored on Anthropic’s servers and used to train future models. Unless you manually opt out. Most users never do.
This is not unique to Anthropic. ChatGPT does it. Gemini does it. Every major AI chatbot trains on user conversations by default. The difference is that with Claude, conversations were not just used for training. They were broadcast to the entire internet.
What to do right now
If you have ever used Claude’s share feature, here are the steps to take. Do them now, not later.
1. Check if your share links are indexed. Go to Google and search "site:claude.ai/share" along with keywords from your conversations. If anything comes up, delete the share link immediately in Claude.
2. Delete old share links. Go through your Claude history and remove any share links you created in the past. Even if they are not indexed now, they could be in the future. Anthropic’s track record on Claude privacy makes this a reasonable precaution.
3. Opt out of data training. In Claude’s settings, find the data training option and turn it off. This does not make your conversations private, but it stops Anthropic from using them to build future models. Every major Claude privacy guide should mention this step, and most do not.
4. Never share sensitive information with any chatbot. This should be obvious by now, but it is not. People put tax documents, medical records, proprietary code, and legal strategies into AI chatbots every day. Treat Claude, ChatGPT, and Gemini like a group chat with unknown participants. Because in a way, that is exactly what they are.
5. Use Claude’s Artifacts carefully. Claude Artifacts are the interactive mini apps and documents you can build inside Claude. These were also affected by the leak. If you built an Artifact with sensitive data, delete the share link and the conversation.
Claude is still a great tool for many things. If you want to get more out of it without paying more, check out our guide on 7 Claude features you are not using. Or if you are building things with Claude, our step-by-step guide to Claude mini apps covers what you need to know about Artifacts.
Claude privacy is not a settled issue. It has broken twice in less than a year. Anthropic clearly knows how to fix it, since they deindexed everything within hours of the story going viral. The real question is why they do not fix it permanently. Until they do, treat your Claude conversations like they could end up on the front page of Google. Because they just did.
