A zero-day vulnerability has been discovered in the Muse smart assistant launched by Meta for macOS users.
The discoverer was Patrick Wardle, a well-known figure in the Mac security community; he named the vulnerability “Not-a-Mused”.
The attack method is not complicated: by altering a hidden audio configuration setting within Muse, it is possible to hijack accounts, obtain authentication tokens, and subsequently gain access to applications associated with Muse, such as email, calendar, and WhatsApp.
The threshold for exploiting such mechanisms is also quite low. A local process can be used to make the changes necessary, or users can be tricked into executing a certain command in the terminal – without the need for complex malware.
Meta has released a hotfix that removes the relevant debugging features. If you are using Muse for Mac, please make sure you have updated to the latest version.
The debugging features have been retained in the official version; such issues are quite common in products that are being developed at a rapid pace. Muse has developed extremely quickly over the past few months, consistently ranking second on the App Store’s free app list – only falling behind TikTok on Thursday. As the number of users increases rapidly, security audits often struggle to keep up.
The vulnerabilities in intelligent agents are far more dangerous than those in ordinary apps.
When an ordinary app is compromised, what gets lost are the data contained within that app. In order to perform tasks on your behalf, intelligent agents possess access to your email, calendar, and communication software – so if one of these tokens is leaked, it’s as if all the doors connecting to these systems have been left unlocked.
This also provides Apple with a perfect excuse for tightening the controls over third-party agents’ permissions. In the beta version of macOS 27.3, the number of steps required to enable computer-controlled use of ChatGPT was increased from four to fourteen. The more such cases arise, the more steps will inevitably be needed.
Before installing any intelligent agent that is capable of operating a computer, make sure you understand what permissions it possesses. No matter how appealing the terminal commands sent by others may sound, never copy and run them directly without first checking them carefully.