Meta has halted the sharing functionality of Muse, its new general-purpose AI agent, following a report from security researchers at Claroc detailing significant privacy vulnerabilities. The feature, which allowed users to generate unique URLs for sharing conversations and other content, inadvertently exposed confidential information to unintended audiences.
Claroc researchers identified a critical design flaw in how Muse constructs shareable links. Unlike typical security practices that rely on obscure, unguessable URL sequences, Muse incorporates the conversation title directly into the domain name structure. For example, share links follow patterns such as “home-sale-budget-9f21c8a3.cf.metaaiusercontent.com,” where the conversation topic is publicly visible. This design choice enables the links to be indexed by passive DNS search services, which record historical domain name resolutions. Researchers demonstrated the vulnerability by using a single passive DNS query to enumerate approximately 6,800 share links created by Muse users and subsequently accessed hundreds of them, discovering sensitive information that was never intended for public distribution.
The vulnerability contradicts Meta CEO Mark Zuckerberg’s recent claims that Muse was “built from the ground up for privacy and security.” Claroc researcher Asaf Aprozper reported the findings through Meta’s bug bounty program on October 8 and attempted to contact Meta personnel directly but received no response. In response to Ars inquiries, Meta acknowledged the report and stated the company is reviewing the sharing mechanism and considering modifications to prevent unauthorized link discovery.
This privacy incident adds to growing concerns about Muse’s security posture. Earlier this year, macOS security expert Patrick Wardle discovered a separate zero-day vulnerability permitting unauthorized applications on the same device to gain complete control of user Muse accounts. Meta downplayed that vulnerability’s severity, though Wardle demonstrated that relatively simple attacks could facilitate account compromise. Given the agent’s requirement for extensive access to user accounts and systems to function properly, the accumulating security issues have raised questions about whether Muse warrants the level of trust required for its operation.
Article Attribution | Read More at Article Source
Article summary produced by Claude AI