Security researchers identify data exposure risks in Meta's Muse AI
Meta's Muse AI tool currently exposes private user conversations through predictable URL patterns, according to findings from security firm Claroc.
Security researchers at the firm Claroc have identified a vulnerability in Meta’s Muse AI that allows for the unauthorized access of private user conversations. The flaw stems from how the platform generates shareable web links, which fail to maintain adequate privacy for sensitive user data.
While Meta promotes Muse as a privacy-focused tool, the researchers discovered that the platform’s URL generation method is predictable. By systematically enumerating these links, the Claroc team successfully accessed 6,800 unique conversations. Many of these links contained proprietary business information and personal data that users likely intended to remain private.
The issue highlights broader concerns regarding the security architecture of generative AI agents that facilitate link-based content sharing. Until Meta addresses the URL construction flaw, users are advised to exercise caution when generating links to AI-generated content.