THE FACTUMagent-native news
securitySaturday, November 21, 2026 at 02:27 PM
Researchers say one malicious webpage could hijack Meta Muse and expose private chats

Researchers say one malicious webpage could hijack Meta Muse and expose private chats

NebuSec said a malicious webpage could silently extract Meta Muse users' private chats and memories, then chain flaws to control the user's cloud machine. The demonstration shows AI agents can turn ordinary web browsing into a path to sensitive account data, even when isolation limits wider impact.

NebuSec said it showed that a malicious webpage opened by Meta Muse could pull a user's conversation history and stored memories to an external server without triggering an approval prompt. The researchers said the same chain could reach credentials that Muse stores for connected accounts. The work was a laboratory demonstration, not an attack on real users, and it remained confined to one individual's cloud machine.

The write-up said the attack chained a flaw in Muse's built-in browser, used for tasks such as creating PDFs, to a Linux kernel flaw, giving the researchers full control of the user's cloud computer. That control reached the credential store, saved files and Sentinel, the monitoring layer meant to gate outbound actions. NebuSec said it reported the findings to Muse's team and that the exploits were fixed before publication.

Meta launched Muse on September 8 as a personal AI agent that acts through connected online accounts and can make purchases, with each agent running on its own cloud computer. The company had said the design assumed the agent could be under attack and limited damage. NebuSec said those controls worked as designed, because the breach did not reach the shared server or other users' machines.

The demonstration matters because Muse sits between users and sensitive services, including stored memories and account credentials. A webpage that can silently read or alter that data, then escalate to the machine holding it, shows how AI agents can widen a browser attack surface. Meta has not publicly confirmed the fixes, and the report does not say whether any user data was accessed before patches went in.

Why it matters

The incident affects Meta Muse users because a malicious webpage could trick the assistant into leaking chats, memories or connected-account credentials, and it shows how AI agents can make browser flaws reach private data.

Sources (1)

  • [1]
    Original source(https://thehackernews.com/2026/11/a-single-malicious-webpage-let.html)