A Single Malicious Webpage Let Researchers Hijack Meta Muse and Access Private Chats
thehackernews.com Nov 21, 2026

A Single Malicious Webpage Let Researchers Hijack Meta Muse and Access Private Chats

AI-summarised brief · reviewed before publication

Researchers at NebuSec demonstrated that a single malicious webpage can hijack Meta’s Muse personal AI assistant, extracting private conversations, stored memories, and even the credentials for connected accounts without displaying any user approval prompt. The attack exploits a flaw in Muse’s built‑in browser, chaining it to a Linux kernel vulnerability that grants full control of the user’s isolated cloud machine. Once inside, the attackers accessed the credential store, saved files, and the Sentinel monitoring layer that is supposed to isolate the agent from sensitive resources. The proof‑of‑concept remained confined to the individual user’s cloud instance and did not spread to shared infrastructure or other users. Meta’s Muse team was notified and patched all identified weaknesses before the findings were published.

💡 Why It Matters

  • · It shows that even tightly sandboxed AI agents can be subverted to steal personal data, exposing a new attack surface for cloud‑based assistants.