Meta has begun rolling out Muse, a personal AI agent that can send email, book travel, make purchases and operate third-party websites. It is available through dedicated iOS and Android apps, Muse.ai and WhatsApp, with support for Meta’s AI glasses planned later. Basic use is free, while heavier automation requires an AI subscription.
Each user’s activity runs in a Secure VM intended to separate untrusted web content from the system that can take actions. A separate control called Sentinel checks information leaving that machine against existing permissions and presents approval prompts directly to the user, rather than routing those prompts through the model. Stripe Link supplies single-use card numbers for purchases.
The current design is not technically inaccessible to Meta, although company policy bars staff access and users can opt out of training. Meta plans a later Confidential VM in which users hold local keys and even Meta cannot enter the environment. It also promises outside audits, published binaries and a transparency log. Until that version arrives, using Muse still requires trusting Meta’s policies alongside its technical isolation.