Meta Suspends Muse AI Sharing Feature Over Privacy Vulnerabilities
Researchers discovered that predictable URL structures exposed sensitive user conversations to public discovery.
1Vulnerability in AI Sharing
Meta suspended the sharing feature for its Muse AI agent following reports that private conversations were publicly accessible. Researchers identified a flaw where shared links could be easily discovered by third parties. This action follows concerns regarding how the platform handles sensitive user data shared through unique web URLs.
2Predictable URL Architecture
The security flaw stemmed from a specific design choice in URL construction. By placing conversation titles directly into the subdomain rather than the URL path, Meta made these links indexable. Researchers from Claroc demonstrated that this structure allowed for the enumeration of approximately 6,800 unique links created by Muse users.
3Passive DNS Enumeration
Discovery of these links was facilitated through passive DNS search services. These systems store historical records of domain name resolutions, capturing data when users access shared links. Because the links utilized the domain cf.metaaiusercontent.com, researchers could easily identify and access content that was never intended for public viewing.
4GhostShares and Data Exposure
Claroc researchers coined the term GhostShares to describe these exposed pages. They found that gathering these links required minimal effort, often taking only a single query to locate. The exposed content included sensitive personal and business information, highlighting the risks associated with automated AI assistant data sharing mechanisms.
5Bug Bounty Reporting Process
Asaf Aprozper, CTO of Claroc, reported these findings through Meta’s bug bounty program on October 8. Despite attempts to reach Meta representatives via LinkedIn, the researchers received no initial response. Meta later acknowledged the report, stating they are currently reviewing the functionality and considering changes to link discovery methods.
6Trust and Platform Access
Muse is designed to autonomously fill forms, book appointments, and manage user accounts, requiring extensive permissions. Meta marketed the tool as a platform built for privacy and security. However, the ease of discovering these shared links raises significant questions regarding the level of trust required for such high-access AI agents.
7History of Security Concerns
This incident follows a previous zero-day vulnerability discovered by Patrick Wardle, which allowed other applications to hijack Muse accounts. While Meta downplayed the severity of that earlier finding, the current link discovery issue further complicates the security narrative. These recurring vulnerabilities challenge the platform's reputation for robust user protection.
8Current Status of Feature
Meta has officially paused the Muse sharing functionality while conducting a comprehensive review of the reported vulnerabilities. Users are currently protected from this specific threat, but the incident underscores the necessity for more secure URL generation practices in AI-driven services. Future updates will likely address these fundamental architectural weaknesses.
Original article: Hacker News — Meta Suspends Muse AI Sharing Feature Over Privacy Vulnerabilities