Particle.news

Meta Launches Muse, an AI Agent That Acts Inside Users’ Apps

The U.S. rollout forces a test of whether deep app access can be kept secure with agents operating on real user data.

Overview

  • Meta released Muse in the United States on Tuesday as a standalone app and as an option inside WhatsApp to perform tasks such as sending email, booking travel, making payments, and controlling smart‑home devices.
  • Muse runs each user agent in its own cloud virtual machine so it can operate in the background and it is designed to connect to specific apps for email, calendar, payments, health, shopping, and home automation.
  • Meta says users choose which apps Muse may access and can revoke permissions at any time and that the company delayed Muse earlier in the year to meet minimum safety, privacy, and performance thresholds.
  • Reporting based on internal Meta tester posts shows mixed reliability and concrete failures, including unexplained disconnections, stopped monitoring of web pages, unauthorized transmissions of private data, and an alleged bypass that exposed iCloud photos.
  • The product’s combination of deep access and autonomous actions heightens privacy and security scrutiny for consumers and regulators and raises the risk that failures could expose sensitive information or disrupt people’s accounts as Meta expands Muse to other devices.