The Register reports that the bug could expose voice prompts, even as Meta promises users control.
A flaw in Meta’s Muse AI app lets local malware redirect dictation traffic, The Register reports. The outlet says the bug could expose users’ voice prompts.
101K
7 Sources, first seen ago
Ars Technica reported that a vulnerability in Meta's Muse assistant for macOS could let a locally running app or a terminal command take over a connected Muse account. The report, based on work by macOS security researcher Patrick Wardle, says Meta released a hotfix after the issue was disclosed.
Muse is built to act across connected services, including email, calendars and social accounts, and it can receive broad permissions on a user's Mac. Ars says Wardle found that any local process could alter undocumented Muse settings, including the endpoint used for dictation transcription. If an attacker redirects that traffic to a server they control, the report says the server can receive the token that authenticates the user's Muse account and use the assistant's existing access.
The flaw was not described as a conventional remote exploit. Ars reported that Meta said as much, but Wardle argued that a ClickFix-style social-engineering trick could be enough to get a victim to run the necessary command. In the reported proof-of-concept scenarios, an attacker could use Muse's privileges to carry out actions such as writing files or accessing connected services rather than needing to build a separate, full-featured macOS stealer.
Not enough discussion yet.
No sentiment analysis available yet.
—
Not ranked yet
—
Not ranked yet
The episode illustrates why the security boundary around an agent matters as much as the permissions a person knowingly grants it. According to Ars, Muse's use of cloud-based transcription made the endpoint setting particularly sensitive; macOS also offers an on-device transcription route. Meta had not explained that design choice in the report. Readers who use Muse on macOS should make sure they have the vendor's latest update and be especially cautious of prompts that ask them to run commands.
Not enough discussion yet.
No sentiment analysis available yet.
