Claude Chats Leaked on Google Search

A significant privacy concern has emerged regarding Anthropic's AI chatbot, Claude, after a large number of shared user conversations and AI-generated documents, known as Artifacts, were discovered to be publicly accessible through Google search. The issue came to light over the weekend when users on Reddit identified shared Claude chats by employing specific Google search operators targeting the site:claude.ai/share domain. Reports indicate that some of these exposed conversations contained highly sensitive personal information, including medical records, confidential company documents, and personally identifiable details of children.

The vulnerability appears to stem from Claude's "share chat" functionality, a feature designed to allow users to generate unique URLs for sharing conversations or projects. While the interface explicitly warns that "Anyone with the link can view," the implication is generally for controlled sharing among trusted individuals, akin to features in services like Google Docs. The unexpected public indexing by Google suggests a broader reach than intended, leading to potential widespread data exposure.

Anthropic, the developer of Claude, initially suggested that user behavior played a role, stating that share links only appear in search results if they are posted on publicly visible platforms like forums or social media. A company spokeswoman, Amie Rotherham, clarified that Claude does not actively share chat directories or sitemaps with search engines and that these links are not discoverable unless users explicitly share them. She emphasized that when a user makes a conversation public, it becomes accessible like any other web content and may be archived by third-party services.

Privacy implications and past incidents

The extent of the exposure is still being assessed, but initial reports paint a concerning picture. A Reddit user first flagged the issue on Saturday, with subsequent reporting by 404 Media on Monday morning detailing various sensitive materials found. These reportedly included detailed patient medical reports, clinical trial results with participant names, documents containing the names and phone numbers of primary school children, internal-use-only company documents, and employee reviews with personal employee data. Exposed Artifacts also contained code snippets and work notes. In one instance highlighted by Fortune, a chat labeled as "shared by Anthropic" reportedly contained AI-generated erotica, a topic explicitly prohibited by Anthropic's usage policy.

This is not the first time Claude chats have been unintentionally made public. Last year, Forbes reported a similar incident where hundreds of Claude conversations were indexed by search engines. While the scale of the current exposure compared to the previous one is not yet confirmed, the method used to find the chats is identical, suggesting a persistent issue with the sharing feature's public accessibility. Furthermore, a 2026 report by 404 Media detailed how a researcher managed to scrape approximately 100,000 ChatGPT conversations that users had elected to share publicly, indicating a broader trend of potential exposure risks with AI chat services.

Google, in response, stated that it does not control what pages become public on the web and that these pages were indexed across multiple search engines. A Google spokesperson, Ned Adriance, noted that search engines provide site owners with controls to manage crawling and indexing, and that these directives are respected. The responsibility, therefore, often falls back on the platform providers to ensure their sharing mechanisms do not inadvertently lead to public data disclosure without explicit user consent beyond the initial sharing action.

To review and manage their shared Claude chats, users are advised to navigate to Settings -> Privacy -> Shared Chats within their Claude account. This allows users to see which conversations have public links and to revoke access if necessary, thereby mitigating further potential exposure.