Science & Tech · Cybersecurity · 1 hr ago
Researchers Demonstrate Attack on Meta Muse That Exposed Private Chats
Researchers demonstrated an attack on Meta Muse, Meta’s personal AI assistant, using a malicious webpage the assistant opened.
The test showed that the page could access a user's private conversations and stored memories without asking for permission.
The researchers then combined flaws in Muse’s browser and the Linux system it ran on to take control of that user’s cloud computer.
That access reached saved files, account credentials and Sentinel, a security layer designed to control what the assistant sends out.
The demonstration stayed on one user's cloud machine and did not reach other users or the shared server beneath it.
The researchers said Muse’s team fixed all the flaws before the findings were published, but Meta has not publicly confirmed the fixes.
The researchers did not say whether anyone’s data was accessed before the fixes were made.
Researchers demonstrated that a malicious webpage could access a Meta Muse user's private conversations and stored memories without an approval prompt.
The attack could also be extended to reach credentials for accounts connected to Muse.
The laboratory demonstration took control of the user's individual cloud machine but did not reach other users' machines or the shared server beneath it.
NebuSec said Muse's team fixed the reported weaknesses before the research was published.
Meta has not publicly confirmed the fixes or commented on the research.
- Who
- Researchers at security firm NebuSec demonstrated the attack.
- What
- A malicious webpage could extract private conversations and stored memories from Meta Muse, and the attack could be extended to reach connected-account credentials.
- When
- The research was published on November 21, 2026. Muse launched on September 8.
- Where
- On the individual cloud machine used by a Muse user.
- Why
- The demonstration showed that vulnerabilities could bypass Muse's protections and expose data and credentials stored on that machine.
This story does not have two clearly opposing sides.
Meta launched Muse, its personal AI agent.
NebuSec said it reported its findings to Muse's team, which fixed the exploits.
The Hacker News published the report on NebuSec's demonstration.
- Security firm
- NebuSec
- Affected assistant
- Meta Muse
- Publication date
- November 21, 2026
- Compromised environment
- One user's cloud machine
- Reported outcome
- Muse's team fixed the weaknesses before publication
What people are saying
So we designed the system to assume the agent may be under attack and limit the potential damage.