Meta Pushes Muse AI Agent for Everyday Tasks, Raising Privacy Concerns

Summary

Meta is promoting Muse, an AI agent designed to carry out everyday tasks such as ordering groceries, contacting customer service, sending emails, booking appointments and planning travel. The system combines a chatbot-like large language model with reinforcement learning to pursue goals set by users, and Meta has presented it as approachable even for people with little technical experience. The article places Muse within Meta’s broader attempt to establish a role in AI: the company spent US$72 billion on AI and other capital expenditure in 2025 and expects that figure to double in 2026, while its Muse Spark 1.2 model reportedly ranks below leading Claude and GPT models. Meta’s strategy is therefore less dependent on having the most capable model than on putting an accessible assistant in front of the billions of people who use its platforms every day. Users can name and customize the agent, decide how often it must request permission, and may eventually interact with it through an avatar or a small standalone device. Muse is currently available only in the United States and connects to email, calendars and selected consumer services; it cannot access bank, medical or tax accounts, although Meta plans to add connectors and expand internationally. The article argues that agents could become a primary interface for online life and give their owners detailed information about users’ preferences, creating both a competitive opportunity and an advertising-related data incentive for Meta. Meta says each agent runs in a dedicated Secure VM and has promised an optional setting under which even Meta cannot access the agent’s contents. The authors remain sceptical because of Meta’s history of privacy controversies and regulatory penalties, and they note that users must actively disable the use of interactions for model training. A journalist also reported that Muse read private messages stored on a Mac. With few rules governing AI agents, the article argues that regulators may lose the opportunity to shape their safeguards if large numbers of people become dependent on them first.