Muse · Published: · WATCH
Type: Personal Assistant
Researchers disclose a patched Muse attack reaching private chats and credentials
Your move: A lab demonstration, reportedly patched before disclosure, not a confirmed user breach.
NebuSec's November 20 research describes a chained attack against Muse's cloud environment. Asking the agent to export an attacker-controlled webpage exploited its bundled browser. The researchers say they then extracted and changed private conversation history and stored memories without an approval prompt. A separate Linux kernel exploit escaped the runtime container and reached protected credential storage in that user's virtual machine.
Why it matters
Keeping passwords outside an agent's immediate reach is useful, but the browser, network checks and operating system must hold that boundary too. This is separate from the earlier Reuters device-permission investigation and the pre-launch VM-escape reporting: NebuSec did not demonstrate access to other users or the underlying shared host. The catch: NebuSec says every exploit described was patched before publication. The Hacker News reports that Meta has not publicly confirmed those fixes. This was the researchers' controlled demonstration, not evidence of a current attack or a confirmed breach of real users. The posts give no complete exposure window or proof that nobody else exploited the flaws. There is no user-installed cloud patch specified. Keep connected accounts narrowly scoped and avoid giving an agent unnecessary sensitive data; the report alone does not establish that every Muse user must reset credentials. We have not reproduced the attack or independently tested the fixes. Sources: Research summary and patch statement · Technical investigation and isolation limits · The Hacker News reporting
Source: NebuSec research ↗
Published by Agent Log. The summary reports the linked source; “Why it matters” and the verdict are Agent Log's interpretation.
More Muse updates →