Meta's AI Agent Muse Found with Zero-Day Vulnerability Allowing Local Code Execution

ME News reports that on September 22 (UTC+8), Beating AI broke the news that just two weeks after its launch, Meta’s personal AI agent Muse was found to have a local zero-day vulnerability by security researcher Patrick Wardle. An attacker only needs to execute a piece of code under the current user’s identity on a Mac to modify a hidden setting in Muse, redirecting voice requests to their own server—without requiring additional macOS permissions. This allows the attacker to intercept what users say to Muse, inject malicious commands, and potentially obtain Muse’s authentication credentials. Muse itself has access to system resources such as files and the camera; any permissions granted by the user to Muse could be exploited by malicious software through Muse. However, this is not a remote exploitation vulnerability—an attacker must first execute code locally on the device. When launching Muse, Meta emphasized its security design, including the Muse Secure VM and a dedicated Sentinel Agent. Yet, the first zero-day vulnerability was discovered in a hidden setting modifiable by a standard local process on the Mac client. (Source: BlockBeats)