Meta has launched Muse, a personal artificial intelligence agent designed to automate digital tasks through a dedicated app, website, and WhatsApp. The service is rolling out to iOS and Android users and is expected to become available through Meta’s AI glasses. Basic access is free, while heavier use will require a paid AI subscription.
Developed by Meta Superintelligence Labs, Muse enters a growing market for agents that can browse the web and operate third-party applications. Meta says users can give instructions in natural language and ask Muse to send emails, arrange travel, sell a vehicle, or complete purchases. Payments are processed through Stripe’s Link system, which creates single-use card numbers and includes purchase protections and no-fee returns.
Meta is emphasizing security and privacy as distinguishing features. Muse initially runs within a Secure VM, a virtual environment intended to separate untrusted web content from the systems authorized to act for users. A monitoring system called Sentinel reviews data leaving the virtual machine and checks whether an action has permission. When approval is unclear, Muse sends a direct confirmation request to the user. Meta says this process is designed to reduce risks including prompt-injection attacks.
The company acknowledges that Secure VM is not technically inaccessible to Meta, although internal policy restricts access. Users may opt out of having their data used for model training.
Meta also plans a Confidential VM option in which users control access keys locally. Developed with Signal creator Moxie Marlinspike, the system is intended to prevent access by Meta or other parties. Independent security firms will audit source code, while binaries and transparency logs will be published for verification.
Muse has undergone internal red-team testing and private bug-bounty reviews. Meta is adding it to its public bounty program, offering rewards of up to $300,000 for qualifying security vulnerabilities.