Meta's Muse Is Better at Surveilling Than Helping Me

Meta’s Muse is a consumer-facing AI agent that automates tasks across Meta’s ecosystem by connecting to email, banking, messaging, and marketplace services. It uses a virtual machine to browse and interact with websites on the user’s behalf, handling actions like ordering food, messaging sellers, and adding items to carts; in one test it ordered a local bakery sandwich (and chose “no tip”). Muse stores long-term “Memory” of preferences and nudges users to link more sensitive data - bank accounts, inboxes, passport and license photos - promising richer automation in exchange. The app achieved rapid early adoption (about 900,000 downloads in its first week) and defaults users into sharing interaction data for model training, with a settings toggle required to opt out. Meta says data is sanitized and plans “confidential” virtual machines later, but details are sparse.

The central finding is that Muse is highly capable but designed to harvest personal data in ways that raise privacy and autonomy concerns. Privacy advocates and researchers warn the default opt-in and constant prompts push users to overshare, and that agentic tools degrade oversight and decision-making, potentially amplifying ad influence and commercial steering. Meta defends data use as necessary for improving models and hints at non-ad revenue plans, yet the experience can make users passive and exposed; the reviewer ultimately deleted Muse after testing, noting the app immediately prompted another connection.