Meta has introduced a new artificial intelligence system called Muse, built to actually carry out tasks for people rather than just answer their questions.
According to the company, the agent can make purchases, book travel, send emails, complete online forms and even negotiate deals on a user’s behalf. Once it is handed a task or a general goal, Muse maps out its own plan and keeps working on it independently, continuing even after the person has closed the app.
The system is designed to check back in whenever conditions change or when it needs a green light from the user, and Meta says it always asks for approval before anything sensitive, such as sending an email or completing a purchase.
For payments, Muse relies on Stripe’s Link service, which can issue a single-use virtual card for an approved transaction so that the agent itself never has direct access to the user’s actual payment information. The assistant is also able to hold onto details a person has shared with it in the past and draw on that memory to make suggestions on its own, without waiting to be asked.
Behind the scenes, Muse runs inside what Meta describes as a Secure VM, essentially its own virtual computer in the cloud complete with a browser, letting it browse websites and interact with connected services much like a person would. Overseeing that access is a second AI system called Sentinel, which decides in real time whether a given action should be allowed to proceed, blocked outright, or sent back to the user for sign-off.
People using Muse can control which services it is allowed to touch and what it can do inside them, and Meta says the agent is kept away from passwords and stored payment credentials, with every action logged in an audit trail users can review.
Meta also addressed privacy concerns directly, stating that anything said to Muse or stored within its virtual machine stays separate from the company’s advertising systems, and that users have the option to opt out of letting their interactions feed into the training of Meta’s AI models. The rollout arrives at a moment when AI companies more broadly are facing scrutiny over the security and privacy risks that come with handing autonomous agents access to people’s accounts and personal data.
For now, Muse is limited to users aged 18 and older in the United States, accessible either through a standalone app or via WhatsApp, with Meta indicating that support for its AI glasses is on the way. Basic use of the assistant is free, while heavier users can opt into paid tiers priced at $20 and $100 a month.
About The Author
