Meta's Muse for macOS Exposed to Serious Vulnerability

On September 25, 2026, a severe zero-day vulnerability was discovered in Meta's Muse for macOS by security researcher Patrick Wardle, who is also the founder of Objective-See Foundation. This flaw allows attackers to hijack user accounts through a hidden configuration option, enabling access to associated applications like email, calendar, and WhatsApp using the obtained system permissions.

Wardle named this vulnerability "Not-a-Mused." Fortunately, following the disclosure, Meta released a hotfix to remove the associated debugging features that contributed to the issue.

The core of the vulnerability lies in Muse's voice input functionality. Unlike Apple's local voice transcription, Muse sends voice data to the cloud for processing. Wardle identified a hidden configuration option named endo_voyager_dictation_endpoint that can be modified by applications or scripts running under user account permissions without triggering additional macOS permission prompts.

Attackers can redirect this configuration option to a server they control, allowing them to intercept user voice commands and account authentication tokens. Once the authentication token is obtained, the attacker can directly manipulate the logged-in Muse account and leverage its existing permissions to read emails, access calendars, or control other connected applications without needing to develop specialized attack software for those applications.

Wardle demonstrated that executing an attack does not require complex intrusion methods. Attackers could use social engineering tactics, such as ClickFix, to trick users into copying and executing a terminal command, thereby achieving initial access and facilitating further attacks.

David Singleton from Meta Superintelligence Labs confirmed that after the vulnerability was disclosed, the company pushed a hotfix update to eliminate the problematic debugging voice configuration option and close this security hole.