Claude exposes private chat leaks into Google and Bing search pages—how to manually turn off the sharing feature?

Hundreds of Claude conversation shares were picked up by Google and Bing and ended up in search results. After the community unexpectedly exposed it, Anthropic and Google started passing the ball back and forth.
（Background: Claude can now help you call Uber, play Spotify.. connects eight everyday life apps; Anthropic is planning the next big move）
（Background info: Anthropic no longer shows summaries of Claude’s thinking process to prevent Fable 5 from being distilled）

Table of contents

Toggle

• One lock, two loopholes
• Who’s responsible?
• How to turn off sharing on your own

When you think only you can see your Claude conversation history, Google may have already indexed it in search results. Over the weekend, a Reddit user found that if you search on Bing for “site:claude.ai/share,” you can surface about 612 results: people asking which political party to join, a Kansas state lawyer asking whether they should proactively disclose when they believe they violated ethics, and even erotic role-play conversations—all laid out in the open for anyone to click.

This isn’t a hack. It’s a textbook-level systemic oversight. Claude lets users create “snapshots,” generating a set of public URLs for sharing a particular conversation with specific recipients. The original intent is convenience, not public posting. Users thought the link was only for friends, but it turned into a public web page that search engines can index for anyone to look up.

One lock, two loopholes

Anthropic, starting at least as early as September 2025, has listed the share pages under /share as off-limits in its robots.txt file, which can be confirmed by Wayback Machine archives. What is robots.txt? Simply put, it’s a publicly accessible list the website places in the backend to tell search-engine crawlers like Google and Bing which pages are “disallowed to read.” This is a long-standing understanding in the internet industry, and in theory, crawlers that see it automatically go around.

But the problem is that this lock only works for crawlers that follow the rules. In its technical documentation, Bing states that developers must, in addition to using robots.txt to block crawling, add a “noindex” tag on each individual web page. What is a noindex tag? In plain terms, it’s a line of instruction in the web page’s code that explicitly tells search engines: even if you crawl it, you must not index it—effectively adding an extra safety layer after locking the door.

Google is even more direct in its developer guidelines: as long as other sites link to a page, even if robots.txt says not to, Google will ignore that directive and still include the page in the index—meaning into the search engine’s database—so everyone can find it later via search, unless the page itself adds a noindex tag, or uses an x-robots-tag in the response headers.

After WIRED reviewed sample pages of leaked Claude conversations, it found that these pages never included a noindex tag from start to finish—locking only the front door while leaving the back door wide open.

The incident spread quickly between July 27 and July 28. Some reports even noted that the leaked content included personal information such as encrypted cryptocurrency wallet keys and names and addresses.

Google search results have been largely cleared for now, but it hasn’t been possible to confirm whether the underlying issue has been fully fixed. Bing search reportedly still has a handful of remaining cases.

Who’s responsible?

An Anthropic spokesperson said the company lets users decide whether to publicly share conversations. They claimed these shareable links can’t be guessed or discovered via search unless the user chooses to share, and emphasized that once a user selects sharing, the content is treated like any other public web page and may be indexed by third-party services. But this sidesteps a key fact: the conversation pages originally didn’t include the noindex tag that could have prevented indexing in the first place, and Anthropic has never directly addressed that.

Google spokesperson Ned Adriance then kicked the ball back. He told WIRED that what pages get indexed is not something Google—or any search engine—can control. Site owners are provided clear tools to decide whether pages can be crawled or indexed, and Google has always respected those settings. In other words: the tools are there, but they weren’t used properly—that’s not my problem. Bing’s parent company, Microsoft, did not respond.

Even more awkward, this isn’t the first time Anthropic fell into the same trap. Back in September last year, Forbes reported that hundreds of Claude conversation records appeared in Google search results. At the time, Anthropic also responded with robots.txt guidance telling crawlers not to access, but as Forbes pointed out then, that has never guaranteed that search engines won’t index specific pages.

Ten months later, the same noindex tag still hadn’t been added. Only after this time’s incident was widely covered by major media did Anthropic issue an additional statement after the report was published—yet it still didn’t explain why the sharing pages had taken so long to get that tag added.

How to turn off sharing on your own

Web version / Desktop steps

• Click the profile avatar at the bottom-left of the Claude screen
• Choose “Settings”
• Switch to the “Privacy” tab
• Find the “Shared chats” item
• Click “Manage” to review all conversations you’ve shared in the list, then press the trash can icon or “Unshare” for each one

iOS / Mobile app steps

• Tap the menu “☰” at the top-left of the screen
• Tap the profile avatar at the bottom-left to enter settings
• Choose “Shared Links”
• Swipe the conversation links you want to turn off from right to left
• Tap the “Delete” button when it appears to cancel public access