Meta announced on Tuesday the launch of Muse, a personal AI agent designed to help people send messages, automate digital tasks, and interact with online services in a secure cloud environment. According to Meta, security and privacy protections are built into Muse from the start.
Meta says Muse is rolling out today to iOS and Android users through a dedicated app and the Muse.ai website. Users can also message Muse directly on WhatsApp and interact with AI agents. People who use Meta AI glasses will soon be able to access Muse through the glasses as well. Muse is free to try, but users who want to automate a larger number of digital tasks will need an eligible AI subscription plan.
Muse is Meta’s latest effort to compete with popular AI agents such as OpenClaw. Users can send natural-language messages to Muse and ask it to complete digital tasks on their behalf. The product was developed by Meta Superintelligence Labs, the division created to help Meta compete with OpenAI and Anthropic. The unit reflects Meta’s belief that AI agents could transform not only its products, but also the way people use the internet.
WIRED previously reported that Meta was testing Muse internally under the codename “Hatch.” Employees used the AI agent to operate third-party applications and browse the web autonomously on their behalf.
In a blog post, Meta said Muse is ready for public use and has been designed with a low learning curve. Users can communicate with the agent in everyday language, while Muse can independently complete tasks such as sending emails, booking trips, and helping sell vehicles.
Muse can also make purchases and complete online checkouts using payment infrastructure developed by Stripe. Stripe’s Link system generates a one-time card number, meaning the AI agent does not need to enter a user’s real payment details across the internet. Meta said Muse is the first AI agent eligible for Link’s agent purchase protection, which includes fee-free returns.
Meta has taken longer than some competitors to release a consumer-facing AI agent, but the company is positioning Muse around security and privacy. In its security and privacy overview, Meta said Muse uses an architecture called Secure VM. The system is designed to isolate each user’s activity inside a virtual machine, separating untrusted web data and third-party integrations from the part of the AI agent that performs actions for the user.
Personal AI agents require significant user trust, particularly when they can access third-party apps, online accounts, and personal information. Meta has faced privacy concerns in the past, making trust a central challenge as it introduces Muse to consumers.
“We know that when you build a product like this that has access to so many sources of personal data, it’s really important to be really responsible for it, so we designed this system very carefully,” said David Singleton, vice president of consumer product engineering at Meta Superintelligence Labs. “And we built something we call Sentinel, which actually monitors everything that moves out of the VM and either matches it with existing policies that the user or the system has allowed for it to happen, or displays a human-involved dialog asking for approval for the action to be taken.”
Source: www.wired.com