- Published on
Muse, Meta's extraordinarily privileged AI assistant, has a serious 0-day
- Authors
- Name
- DeepHorus
Meta's new AI assistant, Muse, has been compromised by a critical zero-day vulnerability, undermining the security promises made by CEO Mark Zuckerberg. The flaw allows locally installed applications and terminal commands on macOS to gain full control over the agent by stealing authentication tokens. This discovery raises significant concerns about the assistant's design, which requires deep integration with user accounts and device resources to function.
Introduced just a few weeks ago, Muse is designed to automate a wide range of personal tasks. It can book appointments, fill out forms, make purchases, generate images, and connect with services like WhatsApp and email. The macOS app requests extensive permissions, including access to the camera, microphone, files, and location data, effectively bypassing Apple's built-in security sandbox to perform its proactive functions.
The vulnerability centers on an undocumented setting that allows any local process to change the transcription endpoint. Normally, voice data is sent to a Meta server for processing. An attacker can exploit this by redirecting the endpoint to their own server, where they intercept the authentication token granting complete control over the user's Muse account and all linked services.
The implications are severe, as the exploit grants attackers the same privileged access the assistant itself holds. This includes the ability to read private messages, access calendars, view documents, and potentially execute purchases or manipulate data across connected platforms. The flaw completely negates the extensive macOS permissions model that Apple spent years developing.
Amazon has already begun blocking Muse from its site, a precautionary measure suggesting broader industry concern over the security risk. Meta has not yet publicly detailed a patch timeline. The incident casts doubt on Meta's claim that Muse was "built from the ground up for privacy and security," especially given the assistant's need for such high-level device access to operate.
The future of Muse now hinges on Meta's swift response to secure the vulnerability. Users are advised to be cautious, as the flaw exists in a live, privileged application. This event serves as a stark reminder of the security challenges inherent in deploying AI assistants with deep system integration and constant connectivity to personal data.
