Stories
Meta Patches Critical Muse AI Flaw Allowing Attacker Control
F
Firstpost•23-09-2026, 01:30
Meta Patches Critical Muse AI Flaw Allowing Attacker Control
- •Meta has issued a patch for its Muse macOS app after a zero-day vulnerability was discovered that could allow attackers to take control of the AI agent.
- •Security researcher Patrick Wardle found that a Muse setting could be exploited to redirect transcription processing from Meta's servers to an attacker's endpoint, granting full control.
- •The flaw stemmed from Meta's design choices, including cloud-based dictation and allowing any app to modify Muse's undocumented settings, some of which control the transcription endpoint.
- •Wardle demonstrated the exploit by taking pictures and writing malicious files via Muse without user alerts, leveraging the AI assistant's privileges.
- •Meta clarified that this was a local privilege escalation attack, not a remote exploit, requiring malicious code already on the user's machine, and a hotfix has been issued.
Key Questions Answered
Security researcher Patrick Wardle found that local processes could change undocumented Muse settings, including the endpoint where transcription occurs, allowing attackers to redirect it to their own server and take full control of the account without triggering user alerts.
Meta developers designed the assistant to allow locally installed apps or executed code to easily modify a list of undocumented settings, and allowed dictation to occur in the cloud rather than on-device.
David Singleton of Meta Labs clarified that this was a local privilege escalation attack rather than a remote exploit, meaning it required malicious code already running on the user's machine, keeping the practical risk quite low.
