
Meta announced it is pausing the conversation-sharing functionality of Muse, the company’s new general-purpose AI assistant, following revelations about potential privacy vulnerabilities in the feature.
Researchers from security firm Claroc reported finding that shared conversations could be discovered with minimal effort due to how Meta constructed the URLs for the sharing links. The firm was able to enumerate approximately 6,800 shared conversation links and access content from hundreds of them that appeared to contain sensitive information never intended for public access. The researchers noted in their report that the discovery process was surprisingly straightforward, requiring just a single query to identify the links.
The vulnerability stems from a design decision in how Muse generates shareable URLs. All shared links use the domain cf.metaaiusercontent.com, and critically, the conversation title is embedded directly in the subdomain portion of the address rather than in the URL path itself. This placement causes the conversation names to be recorded in publicly searchable passive DNS databases whenever someone accesses the shared link. Example links cited in the security research include format variations like home-sale-budget-9f21c8a3.cf.metaaiusercontent.com and q3-revenue-review-7b1e42f0.cf.metaaiusercontent.com. The researchers termed this privacy flaw “GhostShares,” describing pages that become accessible to people who were never provided the link but can discover them through public DNS records.
Claroc’s co-founder Asaf Aprozper reported the findings through Meta’s bug bounty program on October 8 and attempted to contact Meta staff through LinkedIn, though he received no response from either avenue. In response, Meta stated that it had launched the Muse bug bounty program to allow researchers to identify such issues. The company indicated it would review the report and evaluate changes to how these links function.
Muse has achieved rapid adoption following its launch, ranking among the top applications in Apple’s App Store. The AI agent performs tasks including form completion, appointment scheduling, purchases, and document creation, requiring access to most of a user’s accounts. The privacy concern emerges at a time when Muse faces other documented security questions, including a previously reported zero-day vulnerability discovered by security expert Patrick Wardle.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI