Chain of flaws in Meta's Muse AI agent allows full cloud machine takeover in lab demo
Researchers demonstrated a remote code execution chain that compromised Meta's Muse AI agent and its cloud environment without user interaction, accessing private conversations and stored memories. The exploit was confined to a single user's machine, and Muse's team confirmed fixes were implemented before public disclosure.

Researchers chained a browser flaw to a Linux kernel flaw to gain full control of a user's cloud machine in a controlled environment.
Researchers chained a flaw in Muse's built-in browser (used for tasks like creating PDFs) to a Linux kernel flaw to gain full control of the user's cloud machine. The attack demonstrated a chain of vulnerabilities in Meta's Muse AI agent that allowed full control of the user's cloud machine.
Isolated impact within single user environment
The exploit accessed isolated components like credential store and Sentinel without lateral movement to other users or shared servers. The compromised system remained confined to the individual user's machine; no lateral movement to shared servers or other users occurred. The compromise did not extend beyond that one machine; researchers did not break out to the shared server beneath it or into other users' machines.
Flaw fixed before public disclosure
Researchers reported findings to Muse's team and confirmed the exploits were fixed before publication. NebuSec, the security firm that published the research, stated that Muse's team fixed every weakness described in the report. Meta has not publicly confirmed the fixes or commented on the work.
Elementor pro security breach
The malicious webpage in the demonstration accessed conversation history and stored memories and altered them without showing a prompt to the user. Users who connected accounts to Muse should review integrations and rotate credentials if they notice unexpected activity.
Data centre security australia
Muse is Meta's personal AI agent launched on September 8, operating through user-connected online accounts and running on its own cloud computer. Each person's Muse agent runs on its own cloud computer. Muse's design keeps passwords and tokens for connected accounts beyond the agent's reach, with Sentinel as the single gate for anything the agent sends out. The attack did not trigger any approval prompt for the user.
NebuSec is a security firm that published research this week demonstrating a malicious webpage could pull a Meta Muse user's private conversations and stored memories without an approval prompt. The research was a lab demonstration conducted on one person's cloud machine, not an attack on real users. Meta Muse was launched on Sept. 8 as a personal AI agent that works through online accounts and can make purchases on the user's behalf.
Muse's team said: "So we designed the system to assume the agent may be under attack and limit the potential damage." Users should review integrations and rotate credentials if they notice unexpected activity.





