Meta launches Muse, a 24/7 personal AI agent that sends emails and makes payments

Meta officially launched Muse on September 8, a personal AI agent powered by Muse Spark 1.3 that operates around the clock in cloud-based virtual machines, autonomously sending emails, making payments, booking travel, and managing tasks across 300+ connected apps. Distinctive features include nightly reflection and proactive planning — the agent reviews and plans while the user sleeps. Meta offers 100 million free tokens weekly plus $20/month and $100/month premium tiers, accessible via app, web, and WhatsApp.

Mechanically, Meta stresses the agent runs in an isolated cloud environment and never accesses raw passwords or payment details, publishing a deep dive on its safety architecture. Meta AI's official account framed the tradeoff plainly: 'An agent that gets to know you over time necessarily holds a lot of context about you. That's the source of its usefulness and the reason we built' the isolation.

Competitively, Muse is Meta's direct assault on ChatGPT and Claude as personal assistants, leaning on private-VM security as its differentiator. It arrives the same week as Alibaba's Accio e-commerce agent and OpenAI's agentic push, underscoring that 2026 is the year the agent-that-acts-on-your-behalf became the central product battleground.

Skeptics are unconvinced by the privacy pitch. Engineers noted Reuters found unauthorized data exposure in testing and that isolated VMs don't prevent application-layer leaks; the mandatory payment card even for the free tier was read as a data-collection requirement rather than a billing formality. An agent with autonomous payment authority and email access is a high-trust product, and Meta's ad-driven data reputation cuts against the trust it needs. Watch for independent security audits of the VM isolation claims.