Exposed Claude chats on Google include personal data and cryptocurrency keys

A report by 404 Media describes how shareable links to Anthropic’s Claude chats are being indexed by Google, exposing sensitive user content such as an AI-powered therapy app, meeting notes, and a dashboard for analyzing medical billing data. The exposed chats also reportedly include private cryptocurrency wallet keys and personal information like addresses, according to the report.

The issue appears tied to a user-controlled data-sharing setting. Anthropic responded that it does not share chat directories or sitemaps with search engines and that shareable links are not discoverable unless users choose to share them. The company stated that when users share a conversation, they make that content publicly accessible, akin to other public web content that may be archived by third parties.

The debate centers on whether Anthropic’s privacy principles adequately address the risks of user-generated shareable links. Critics argue that certain content patterns—such as “vibe-coding” or repetitive plaintext—can inadvertently reveal sensitive information, making it easier for attackers to infer or reconstruct private data even without direct access to the chat.

The exposure does not stem from a platform-level breach but from user-created shareable links, shifting responsibility to individual users while raising broader questions about default data-sharing defaults and the long-term privacy implications of AI chat platforms.