Skip to page content

AEGLIX MIND

Connected intelligence.Security research, connected

AEGLIX news · Global

Researchers Demonstrate Full Takeover of Meta Muse Cloud Instance via Chained Browser and Kernel Flaws

Based on reporting from The Hacker News

Published by the source: Sat, 21 Nov 2026 17:23:07 +0530

AEGLIX News

Security researchers at NebuSec demonstrated that a single malicious webpage could extract private conversations, stored memories, and account credentials from a Meta Muse user's dedicated cloud machine without triggering any approval prompt, the firm disclosed yesterday. The laboratory demonstration chained a flaw in Muse's built-in browser to a Linux kernel vulnerability, achieving full control of the individual user's cloud computer and reaching components the security architecture specifically isolates, including the credential store and the Sentinel monitoring layer designed to gate all outbound agent actions. NebuSec stated the compromise remained confined to the targeted user's cloud instance and did not extend to the underlying shared server or other users' machines.

Meta launched Muse on September 8 as a personal AI agent that operates through connected online accounts and can make purchases on a user's behalf. Each user's agent runs on an isolated cloud computer protected by Sentinel, a monitoring layer intended to serve as the single gateway for anything the agent sends externally. The research targeted this architecture by exploiting the agent's need to browse the web, using a malicious page to break out of the browser sandbox and then leveraging a kernel flaw to take over the entire virtual machine, thereby bypassing the walls separating the agent from stored credentials, saved files, and Sentinel itself.

The attack chain began when the Muse agent visited a malicious webpage, which exploited a browser vulnerability to execute code and then a Linux kernel flaw to escalate privileges to full machine control. With that control, the researchers could pull the user's conversation history and stored memories to an external server, alter that data, and access the credential store for connected accounts. Critically, no user prompt appeared during the extraction or modification. The researchers emphasized the work was a controlled laboratory demonstration against their own instance, not an attack on real users, and confirmed they did not break out to the shared infrastructure or other tenants.

According to NebuSec, the vulnerabilities were disclosed to Meta's Muse team and all exploits had been remediated prior to the research being published. The security firm states that Muse's engineers have since addressed every weakness outlined in their analysis. Meta has not issued any public confirmation of these fixes nor commented on the findings, and the published write-up does not indicate whether any user data was compromised before the patches were deployed.

Report prepared 2026-11-21 · Based on reporting from The Hacker News. Source publication date above; subsequent developments may not be reflected.

Sources

Read original source · The Hacker News (opens in a new tab)

Privacy policy

Explore the story

Consult the cited publisher for the original reporting, updates and supporting evidence. Check the publication date and source before acting on a headline.

These are research tools; no CVE, indicator, or campaign relationship is inferred from this article.

More from The Hacker News

← Back to security news