Meta patches Muse exploit that let attackers control the AI agent

Meta has patched a security vulnerability in its Muse macOS app that allowed attackers with local access to a user's device to take control of the AI agent and access the user's account. The exploit worked by redirecting the app's cloud-based transcription processing to an attacker's own server, and attackers could use the compromised agent to take pictures and write files to disk. The vulnerability existed because of several design choices, including performing dictation in the cloud rather than on-device and allowing any app to control undocumented Muse settings. While Meta stated the real-world risk was minimal since the exploit required local device access, a security researcher noted that the company should have prioritized security from the beginning of development.