Last weekend, users discovered that Claude's AI session sharing appeared directly in Google search results. The core issue is that Anthropic did not set a no-index tag on the share links, so as long as the SITE: command is used, a large number of session sharing links already indexed by Google or other search engines can be found, some of which even contain users' private information.

It should be clarified first that this is neither a breach of Anthropic's backend database nor a leak of chat records that users never shared. What was crawled by search engines were mainly session sharing links generated by users themselves - these links can be accessed publicly by anyone who knows the address. The problem is that many users thought the sharing links were only for colleagues or friends, but they didn't realize that as long as the link appears anywhere on public web pages, forums, social platforms, etc., it could be discovered and indexed by search engines, allowing strangers to find the link through search and directly view the conversation content.

image.png

After testing, some users found that their session shares contained wallet private keys, illegal consultations, company task information, and identifiable personal content. Such information being exposed on the Internet is clearly a huge security risk. Although Claude warns about the risks when users generate a share link, many users are unaware that once a share link is created, there is a possibility of the session being leaked, and sensitive content inside may be seen by others publicly.

More users might think that as long as they don't forward the link to strangers, it's safe - non-professional users obviously don't understand how search engine crawlers work. Thus, Anthropic should have set a no-index tag, but due to improper configuration, a large number of share links were successfully crawled by Google, Bing, Yandex, and other search engines.

Currently, Anthropic has communicated with search engine providers to remove the links, but some engines may still have residual or cached links; if users have shared sessions containing sensitive content, it is recommended to find that session and delete it directly.

The root cause of this problem is likely a technical configuration error. The /share/* path of Claude was set to be disallowed in robots.txt, but the page response header also added the X-Robots-Tag: none attribute, which is equivalent to noindex and nofollow. It sounds like double insurance, but the actual effect may be counterproductive. Google's official documentation clearly states: if a page is blocked from crawling by robots.txt, search crawlers cannot read the noindex or the X-Robots-Tag in the response header. As a result, search engines may not be able to crawl the main content, but they may still discover this URL through external links and display it in the search results. Thus, a large number of URLs under claude.ai/share/* were crawled and indexed by Google search. Although the page description shows "No available information," other users can still click on the link and directly view the session content.