Meta Muse AI Assistant Zero-Day: What Security Flaws Mean

Secondo Ars Technica, Meta's new macOS AI assistant, Muse, contains a serious zero-day vulnerability that lets locally installed apps bypass system restrictions. Security researcher Patrick Wardle found that any local code can change undocumented settings, specifically the endpoint where voice transcription occurs.

How does the Muse zero-day compromise security?

Meta designed Muse so that unprivileged local apps can modify system settings without standard macOS security checks. Attackers can exploit this by redirecting the transcription server address to their own endpoint, capturing the authentication token and gaining full control over the user's Muse account.

What are the practical consequences for users?

Because Muse integrates deeply with WhatsApp, email, calendars, and payment tools while requesting broad macOS permissions, a hijacked assistant gives attackers a ready-made malware vector. Instead of writing complex malware, attackers can leverage the AI agent itself to read files or dispatch private messages.

The fallout extends beyond security exploits. Amazon began blocking Muse from its shopping platform roughly 12 hours before public disclosure, classifying it as an unauthorized agent that violates operational conditions.