Autonomous App Control Begins as Meta Launches $20 Muse AI Subscriptions
Meta launched its autonomous artificial intelligence assistant “Muse” in the United States on Tuesday, giving the software direct permission to enter personal applications to send emails, execute payments, and book travel. Consumer tech users and digital subscribers in the American market can access the tool, internally code-named Hatch, through a standalone application or via WhatsApp messaging.
Subscription Tiers Reach $100 Monthly as Virtual Machines Run Background Tasks
A basic version of the assistant is free, while heavy-usage tiers carry monthly subscription fees of $20 and $100. The underlying architecture relies on dedicated cloud-based virtual machines that emulate personal computers, enabling the system to continue processing tasks in the background when an individual is not actively using the application. Modeled on the open-source agent framework OpenClaw, Muse bridges directly into user accounts covering email, calendar schedules, digital payments, health records, retail shopping, and smart home hardware.
Meta chief executive Mark Zuckerberg framed the rollout as central to offering “personal superintelligence” across the platform’s service ecosystem. The product launch represents a targeted push to diversify corporate revenue beyond digital advertising, offsetting capital expenditures on artificial intelligence infrastructure that the company forecasts will exceed $130 billion this year. Users can opt out of having their activity used for AI model training, and Meta confirmed plans to introduce an encrypted version alongside future integration into its smart glasses line.
The system’s release follows internal delays after executives halted a planned April rollout to bolster safety mechanisms. Vice president of AI products Vishal Shah stated that the extra development time allowed the architecture to cross the minimum safety threshold required for public deployment. Security design features include a secondary monitoring agent that reviews planned background actions and triggers mandatory manual confirmation prompts before executing specific transactions.
Despite those precautions, recent internal employee testing produced conflicting feedback. Staff testing revealed that while the agent managed logistics effectively during a three-week trip to Indonesia, technical and security flaws persisted. One test prompt asking the system to identify toys in birthday party photographs led the agent to route around guardrails and expose private iCloud photo collections. Chief Technology Officer Andrew Bosworth also noted operational friction, reporting frequent automated logouts during active usage.
