Meta Muse Exposes Privacy Vulnerability: Over 6,800 Chat Links Found via Search, Corporate Secrets Also Exposed

Beating AI Express: US cybersecurity firm Claroc has discovered that chat share links for Meta’s personal AI agent Muse can be bulk-retrieved by strangers. Researchers found over 6,800 share links after a single query to a DNS history database. After sampling hundreds of links, dozens of sensitive documents were uncovered, including internal corporate revenue reports, small business loan applications, and preparation materials for a candidate’s secret recruitment communication with a leading AI lab—even internal contacts and meeting times were exposed. When users share chats via Muse, the system generates a viewable webpage requiring no login, embedding the chat title and a random ID into the domain. Once someone opens the page, the full domain may be logged by third-party DNS services. Claroc directly searched these historical records to obtain share addresses that would otherwise be hard to guess. While Muse requests search engines not to crawl these pages, it cannot prevent domains from being logged by other services. Claroc reported the vulnerability to Meta as early as October 8 and published its research on October 18. It was not until October 20, after US tech media Ars Technica intervened to inquire, that Meta confirmed it had suspended the chat sharing feature. On the same day, Claroc rechecked 40 links that had previously been accessible without login and found they were all invalid, though the chat titles remained in DNS history records. Last month, Meta CEO Mark Zuckerberg had claimed that Muse was a product “designed from the ground up for privacy and security.”