
Security firm NebuSec has published research showing how a malicious webpage could be used against Meta Muse, Meta’s cloud-hosted personal AI agent. According to NebuSec, the attack 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 inside the agent’s runtime container. From there, weaknesses in how Sentinel, Muse’s outbound-traffic monitor, classified and inspected requests let the researchers send conversation history and stored memories to an external server without triggering an approval prompt, and also alter those memories.
The researchers then chained a flaw in the shared Linux kernel to escape the container and gain root in the user’s own cloud VM, reaching the secure credential store, saved files and Sentinel itself. NebuSec says its exploit-generation framework succeeded in more than 99.9% of test runs. The work was a lab demonstration rather than an attack on real users, and The Hacker News reports it stayed inside one user’s machine: the researchers did not reach the shared host or other users’ machines. NebuSec says it reported the findings to the Muse team and that all the exploits described had been patched by the time of publication. Meta has not publicly confirmed the fixes or commented, per The Hacker News, and the write-up does not say whether any user data was accessed before the patches.
The research is a case study in layered defenses failing together: Muse was designed to assume its agent might be turned against the user, with container isolation, network approval and separated credentials, and NebuSec’s point is that a chain of weaknesses across the browser, the network monitor and the kernel got through all three. NebuSec is a security firm that sells code-security and audit services, and the write-up credits its own Automated Exploit Generation framework. The Hacker News notes that this is separate from an earlier Muse flaw found in September, which only let malware already on a user’s Mac hijack the assistant on the device.
Sources
- NebuSec: Hacking Muse: Stealing Private Chats Through Browser Use
- NebuSec: Inside Meta Muse: Browser, Sentinel and the Shared-Kernel Boundary
- Muse: How We Built Safety Into Muse (Muse’s launch safety post, linked from The Hacker News; not read in full)
- The Hacker News: A Single Malicious Webpage Let Researchers Hijack Meta Muse and Access Private Chats
