Claude share links briefly made AI chats searchable on Google

Security researchers and users found that hundreds of Claude conversations became discoverable in Google after public share links were indexed. The links led to pages that contained legal advice, engineering work and private discussions.

The shared pages were created when users generated public URLs through Claude’s sharing feature. When those URLs were posted in public forums or social feeds, search engines were able to index them. Many listings have since been removed from search results, but saved bookmarks, archived copies or links posted elsewhere may remain available unless Anthropic revokes them on its servers.

The exposure involved the optional share links rather than private chats. The shared pages did not appear to include protections that prevented indexing, allowing copies of conversations to appear in search results until listings were deindexed or removed.

Professionals use Claude for tasks that involve sensitive material, including contract review, debugging proprietary code, analysis of business documents and organizing financial records. One user reported uploading years of files while recovering a lost bitcoin wallet containing five BTC, illustrating the type of confidential information that can end up in chat histories or shared links.

Anthropic’s consumer terms state that conversations may be retained under certain conditions and describe options for model training choices and data settings. The company acknowledged the presence of public share links and described the controls users have to manage shared conversations and data settings.

Researchers have shown how AI agent attack techniques can manipulate tools that have access to browsers, email, cloud storage and financial services. Separately, a malware campaign known as Openclaw impersonated AI software to harvest passwords and cryptocurrency wallet credentials, demonstrating an avenue threat actors use to target users of AI interfaces.

U.S. regulators are seeking information on chatbot data practices. The Federal Trade Commission issued guidance on privacy and confidentiality for generative AI and opened an inquiry into how chatbots collect, retain and protect data. Regulators requested details about retention policies, user protections and internal safeguards.

Users who created Claude share links are advised to review active shared conversations in the platform’s privacy settings and remove links that are no longer needed or that contain confidential material. Security researchers recommend avoiding posting public URLs with sensitive information, limiting AI access to credentials and critical files, and regularly reviewing permissions granted to AI applications that integrate with personal or business data.