A major security slip exposed an unknown number of Claude shared chats and interactive Artifacts to the entire web. Over the weekend, Reddit users discovered that using simple Google search operators, like site:claude.ai/share, brought up thousands of private conversations. Google search results indexed sensitive personal information, including real medical reports, confidential corporate strategy files, internal employee performance reviews, and phone numbers belonging to young children.

The issue stems from the share chat feature inside the Claude interface. When users click this button to create a public link, Anthropic generates a unique web address. The interface warns users that anyone with the link can view the page, implying that the conversation is meant for selective sharing with friends, coworkers, or small groups. However, search engine crawlers scraped these web addresses and added them directly to public search indexes.

Anthropic shifted initial blame onto users who posted their shared links publicly on internet forums and social media channels. The company explained that search engine bots automatically crawl and index links published on open web pages. Anthropic stated that it does not submit sitemaps or user directories directly to search engines. Representatives clarified that when a user posts a link on a public site, web crawlers treat that link like standard public content.

Despite those statements, journalists testing the breach found sensitive files that nobody intended to share with the public. Investigators uncovered a detailed medical report containing real patient names, internal company documents marked for private office use, and custom code blocks. Reports also noted that a search surfaced an Artifact labeled “shared by Anthropic” that contained explicit erotic stories generated by the model.

Google addressed the indexing situation by pointing to web crawling standards. A spokesperson confirmed that search engines index pages whenever links appear on public sites unless site owners set up explicit rules to block crawlers. Google provides clear webmaster controls, like canonical tags and robots.txt directives, allowing platform developers to keep shared links out of search engine databases.

This incident marks the second time Anthropic faced a major indexing breach. Last year, search engines indexed hundreds of private Claude chats before technical teams pulled the pages down. Similar privacy failures have hit competing AI platforms in the past, including an incident where researchers scraped over one hundred thousand shared ChatGPT conversations indexed by search engines.

Anthropic took steps to clean up the current exposure, and standard search operator queries no longer display exposed pages. However, the leak serves as a harsh reminder for anyone using cloud software tools. If you generate a shareable web link for an AI conversation, treat that link as public data. To check which Claude chats you set to public, navigate to your account settings, select privacy, and review your shared chat history.