Meta Suspends Muse Sharing Feature After 'GhostShares' Leak
AI-generated image · Bay Street Wire
Security researchers discovered that a design flaw in Meta's AI agent allowed the public discovery of thousands of private user conversations.
Meta has suspended the sharing functionality of its Muse AI agent after researchers discovered a flaw that exposed sensitive user data, as first reported by Ars Technica. Researchers at the security firm Claroc identified a vulnerability they termed "GhostShares," which allowed them to enumerate 6,800 links created by users to share conversations.
Claroc researchers found that Meta's decision to place conversation titles within the subdomain—rather than the URL path—made the links discoverable via passive DNS searches. By utilizing commercial services that track DNS records, researchers were able to uncover private information, including a national marketing company's internal revenue review for its CEO, a small-business owner's loan application, and a real-estate investor's performance report detailing mortgage and income figures. Other exposed data included a candidate's confidential executive-search notes and a workplace vacation calendar.
Asaf Aprozper, co-founder and CTO of Claroc, stated he reported the findings through Meta's bug bounty program on October 8. In a statement provided to Ars Technica, Meta confirmed it has paused the feature to review the report and consider changes to how links are discovered.
This incident follows other security concerns regarding Muse. Ars Technica reports that macOS security expert Patrick Wardle previously identified a zero-day vulnerability that could allow apps on a device to seize control of a Muse account, a finding Meta downplayed by stating exploitation required a previously compromised device.