Particle.news

Meta Launches Muse, a Personal AI Agent That Can Access Users’ Apps

The U.S. release tests whether Meta’s claim that each agent runs in a dedicated virtual machine will reassure users after internal testers reported reliability and data‑exposure problems.

Overview

  • Meta launched Muse in the United States on Tuesday, Sept. 8, 2026, with a web app, iOS and Android apps, WhatsApp integration, a free tier that requires a payment card, and two paid plans at $20 and $100 per month.
  • Muse is built to connect to a user’s email, calendar, payments, shopping, health and smart‑home services so it can autonomously send emails, book travel, make purchases and complete multi‑step tasks using browser access or APIs.
  • Meta says each Muse agent runs in its own Muse Secure VM that prevents the agent from seeing passwords or payment methods and keeps conversations out of the company’s ad systems, a claim that Meta says improves privacy and background tasking.
  • Internal Meta tester posts and Reuters reporting describe reliability and security failures such as unexplained disconnects, frequent forced logins, silent task failures, and cases where an agent exposed or uploaded sensitive personal data like iCloud photos.
  • The launch follows a roughly $18 billion multistate settlement and a history of high‑profile privacy cases, raising the likelihood of close regulatory scrutiny and affecting whether users will trust Meta to give an AI deep access to their personal accounts.