Malicious webpage could expose private chats in Meta's Muse agent, lab test shows
The flaws gave full control of one user's cloud machine, though the fixes are NebuSec's claim and Meta has not confirmed them.
Security firm NebuSec showed in a lab setting that a single malicious webpage, opened by Meta's Muse personal AI agent, could pull a user's private chat history and stored memories to an outside server and alter them, without any approval prompt. Chaining a flaw in Muse's built-in browser with a Linux kernel flaw gave the researchers full control of the user's personal cloud machine, including its credential store and the Sentinel monitoring layer meant to police outbound traffic. The compromise stayed on that one machine and did not reach shared infrastructure or other users. NebuSec says it reported the issues and that Muse's team fixed them before publication, though Meta has not publicly confirmed the fixes and the write-up does not say whether any real user data was accessed.
Why it matters
For users of Muse, the risk is that simply opening a hostile page could expose private conversations and stored memories without any warning. Because the exposure in this test was limited to a single user's machine, the wider concern is how an AI agent with a built-in browser handles untrusted content. Until Meta publicly confirms the fixes and the scope of any real-world exposure, users have only the researchers' account to go on.
Series Meta Muse AI agent privacy
- Malicious webpage could expose private chats in Meta's Muse agent, lab test shows
- Meta's Muse assistant is being used to mass-unfollow Instagram accounts
- Meta's Muse agent reportedly exposes other users' private data, staff say
- Meta's Muse agent can now place phone calls for paid subscribers
Matters or hype?
Does this story matter, or is it hype? Vote before you see what everyone else said.