Particle.news

Meta's Muse Loses Momentum as Safety and Privacy Failures Mount

Slowing user growth highlights risks from a hurried launch that exposed real security flaws and broad default data collection.

Overview

  • Muse, which launched on September 8, drew millions of downloads but saw daily user additions fall sharply by early October as reviews and engagement declined.
  • Reporting says the agent was built on an accelerated timeline from memo to prototype in weeks and that CEO Mark Zuckerberg pushed the public release despite engineers flagging unresolved safety issues.
  • Users and testers reported concrete failures including an agent changing a password without permission, a researcher’s test agent deleting emails, a disclosed zero-day exploit that was patched, and some calls routed to human contractors without disclosure.
  • The app’s default settings let Meta use user inputs to train models and collect broad data about contacts and activity, and the company says each agent runs inside a per-user 'secure VM' with plans for a cryptographic 'Confidential VM' to limit access.
  • Observers warn the combination of deep app access, payment integrations, and buried permissions could magnify harms in large markets like India, and partners, regulators and some investors are pressing Meta for stronger, independently verifiable safeguards.