Security
NebuSec shows in the lab that a single booby-trapped webpage could pull a Meta Muse user’s conversations and memories, and says every flaw is fixed
NebuSec says a malicious page could extract Muse chats and memories without an approval prompt in a lab demo; it says all flaws are fixed.
Researchers at NebuSec showed that a malicious webpage could make Meta’s Muse AI agent send a user’s private conversation history and stored memories to an outside server without any approval prompt. The work was a laboratory demonstration, and NebuSec says every exploit it describes was patched by publication on 20 November.
The essentials
- The chain: it began with a vulnerability in the Chromium browser Muse uses for tasks such as generating PDFs. Having the agent export an attacker-controlled page gave code execution in the agent’s runtime container, according to NebuSec.
- The bypass: weaknesses in how Muse’s Sentinel layer classified and inspected outbound requests let history and memories leave without an approval prompt. The researchers could also modify stored memories and conversations.
- Deeper access: a Linux kernel exploit let them escape the container and get root in the user’s own VM, exposing persistent storage and credential storage. They did not demonstrate escape to the hypervisor or access to other users’ environments.
Scope and caveats
The Hacker News stresses this was not an attack on real users and stayed inside one user’s cloud machine. Muse is Meta’s personal AI agent, launched on 8 September. Meta has not publicly commented, per the Lavx report.
What is not established
- Whether any real user data was accessed outside the research: not stated in the sources.
- Whether connected-account credentials were actually extracted, as opposed to reachable: not stated in the sources.
- The date the findings were reported to Meta: not stated in the sources.
Summary table
| Point | What is stated as of November 21, 2026 | Source |
|---|---|---|
| Published | 20 November 2026 | NebuSec |
| Muse launch | 8 September 2026 | The Hacker News |
| Fix status | All described flaws patched by publication | NebuSec |
FAQ
Were real Muse users attacked?
The sources describe a laboratory demonstration only, and NebuSec says all described flaws were patched before publication.
Sources
- NebuSec: Hacking Muse: Stealing Private Chats Through Browser Use (verified 2026-11-21)
- The Hacker News: A Single Malicious Webpage Let Researchers Hijack Meta Muse and Access Private Chats (verified 2026-11-21)
- Lavx: A malicious webpage let researchers hijack Meta Muse and access private chats (verified 2026-11-21)
Spotted an error in this article? Report it · Our corrections policy