Meta has temporarily suspended the sharing feature in Muse, its general-purpose AI assistant, following a security report published by Claroc researchers that identified a method for accessing shared conversations without authorization.
The vulnerability centers on how Muse constructs URLs for shared content. Like many online services, Muse allows users to create shareable links to conversations and other materials. While these links typically incorporate long, unpredictable character sequences to maintain privacy, Claroc researchers discovered that Muse’s URL structure makes enumeration straightforward. The core issue stems from the decision to include conversation titles directly in the domain name rather than hiding them in the URL path. This design choice causes the information to be recorded in passive DNS databases—searchable repositories that log historical domain name resolutions. Using a single passive DNS query, the researchers were able to discover approximately 6,800 share links created by Muse users.
After accessing a sample of these links, the researchers encountered substantial quantities of confidential information never intended for public disclosure. According to Claroc, the vulnerability enables what they term “GhostShares”—pages accessible to individuals who never received the direct link but can discover and open them through DNS searches. The researchers noted that discovering these URLs required minimal effort, contrary to their initial expectations.
Claroc co-founder Asaf Aprozper reported the findings through Meta’s bug bounty program on October 8, though he stated that neither the official channel nor direct LinkedIn outreach generated responses. In a statement to Ars Technica, Meta acknowledged the report and confirmed it has paused the sharing functionality while reviewing the matter and considering modifications to how such links can be discovered.
This incident represents the second notable security concern affecting Muse since its recent launch. Earlier, macOS researcher Patrick Wardle identified a separate zero-day vulnerability permitting other applications on the same device to assume control of user accounts. The timing of these issues has raised questions about whether the platform merits the extensive access to user accounts and personal systems that its autonomous capabilities require.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI