Capabilities and vulnerabilities of the Muse AI agent | Kaspersky official blog
Source Kaspersky Security Blog
Rated High by Threat Radar's AI analysis
Data as of 12 Nov 2026, 17:04 UTC
What to do
AI-suggested
Meta has released an update that fixes the zero-day vulnerability allowing server redirection and remote control of Muse. Users should ensure their Muse app is updated to the latest version. Additionally, users should review and restrict Muse's permissions, particularly disabling Full Disk Access, Automation, and unnecessary app connectors. Setting action confirmation to 'Always ask' or 'Ask for some actions' limits the agent's autonomy. Users should also be cautious about executing commands in Terminal and avoid installing untrusted software or browser extensions that could alter system settings. Monitoring and adjusting permissions regularly is recommended to maintain privacy and security.
Description
Muse is Meta's AI agent available on iOS, Android, macOS, and web, designed as a personal assistant with extensive access to user data and device functions. Shortly after launch, it was found to access user conversations without explicit permission, likely due to opaque permission controls. A zero-day vulnerability was discovered allowing attackers to redirect Muse's voice recognition server to a malicious endpoint, enabling remote control of the agent with the victim's authorization token. This could allow attackers to access camera, messages, files, and other sensitive data, and perform actions on behalf of the user. Meta has released an update fixing this vulnerability. Additionally, an experimental feature involving human operators handling calls was discontinued after privacy concerns. Users are advised to review and restrict Muse's permissions to mitigate risks.