Zero-Day Exposes Meta's Muse AI Assistant to Total Hijack

Meta's new AI assistant Muse faces serious security scrutiny after researchers disclose a zero-day vulnerability that allows any locally installed app or terminal command to take complete control of the agent. The flaw undermines CEO Mark Zuckerberg's claims that Muse is "built from the ground up for privacy and security." Adding to the concerns, Amazon has begun blocking Muse from its site.

Muse, which Meta introduced a few weeks ago, books appointments, fills out forms, makes purchases, and connects to a user's WhatsApp, email, calendar, and social media accounts. To function, the macOS app requires authentication to each of these services and broad permissions to device resources like the microphone, camera, location, and disk. These are exactly the protections Apple has spent years building into macOS to protect users from untrusted apps, and Muse effectively undoes them.

The vulnerability stems from Meta's design, which lets any local process, regardless of its own macOS permissions, access the token authenticating users to their Muse account and change a long list of undocumented settings. While most settings are harmless, one allows attackers to change the endpoint where transcription occurs, redirecting sensitive data to a server they control. A simple ClickFix attack is just one of several ways researchers demonstrate the flaw can be exploited.

Read More at the original source →