Meta Unveils Muse, Its AI Agent, With So Many Privacy Questions

Meta has unveiled Muse, a personal artificial intelligence agent that can carry out everyday tasks on behalf of users, making its most ambitious push yet into consumer AI while raising new questions about privacy and trust.

The launch comes less than two weeks after Meta agreed to an $17 billion multistate settlement over allegations concerning the consumer harms associated with its social media platforms, putting the company's handling of personal data under renewed scrutiny.

Muse is designed to go beyond conventional AI chatbots, which largely answer questions, generate content, and act as conversational assistants. The new agent is intended to take action across the services people use in their daily lives.

To operate Muse, users can connect individual apps and services, including email, calendars and payments, as well as health and fitness platforms, smart-home systems, dining, shopping, music and event services.

The agent can then perform tasks such as sending emails, booking travel, negotiating lower bills, completing forms, creating plans, turning recipe videos into grocery lists, sending invitations and making purchases.

For transactions, Meta said Muse will use Link by Stripe for checkout. The integration includes purchase protections, a feature that could help address concerns over allowing an AI system to make purchases on a user's behalf. Meta said integrations with Shopify's Shop Pay and 1Password are also planned.

The launch marks Meta's entry into a rapidly developing market for so-called agentic AI, in which software is designed not simply to respond to instructions but to execute multi-step tasks with limited human intervention.

Companies ranging from major technology firms to startups are pursuing similar systems, including AI-powered browsers and assistants capable of taking actions for users. Others are embedding agentic capabilities into communication platforms that consumers already use.

Meta is positioning Muse around user choice. Consumers can connect apps and services individually rather than granting the agent access to everything at once. If a service is not supported by one of Muse's built-in connectors but provides a public API, Muse can establish a connection using credentials supplied by the user. Where an API is unavailable, the agent can interact with the service through a browser.

Muse is powered by Meta's Muse Spark AI model and operates inside what the company calls Muse Secure VM, a dedicated virtual machine with its own browser.

Meta said the system has been designed with separate privacy and security controls. A second system, called Sentinel, operates on the same virtual machine but is isolated from Muse at the system level. The company said Muse will not have access to users' passwords or payment methods. Meta also said conversations and other data generated through Muse will not be shared with its advertising systems.

The safeguards are necessary because an AI agent with access to email, calendars, shopping accounts, financial services and other applications potentially has a much broader view of a person's life than a conventional chatbot. That creates a different trust equation for consumer AI. A chatbot that produces an inaccurate answer can be corrected or ignored. An agent that can send an email, make a purchase, or change a service on a user's behalf can create real-world consequences if it misunderstands an instruction or acts on incorrect information.

Privacy concerns have already emerged around other agentic AI products. Early users of the AI assistant Instinct, for example, raised concerns after discovering broad language in its terms that granted the service a "perpetual and irrevocable" license covering user materials, including rights related to access, storage, distribution, and AI-model training.

Meta's approach is therefore built around limiting access and separating the agent from sensitive authentication information. The company is betting that those controls will give consumers enough confidence to delegate consequential tasks to Muse.

The service will initially be available through the web at muse.ai, as well as iOS and Android applications and chats in WhatsApp. Meta said Muse will also eventually be available through its AI glasses.

Muse will be free initially, although Meta plans to introduce usage limits and paid subscriptions. The company said most users are expected to remain on the free tier. At launch, the paid Power plan will cost $20 a month, while the Maximum plan will cost $100 a month. Both plans provide greater usage capacity for handing everyday tasks to the agent.

Users will be able to monitor their remaining usage through a meter in the app. Meta said users will receive warnings when their free allowance is exhausted and will then have the option to subscribe.

Muse is also designed to continue working after users leave the application. Meta said the system can learn from conversations about what matters to individual users and eventually make suggestions without being explicitly prompted. That feature could make the product more useful, but it also increases the amount of personal information the system may process over time. The more an agent learns about a person's preferences, schedules, and routines, the more useful its recommendations can become, but the greater the consequences if that information is mishandled.

Meta's challenge, therefore, extends beyond building a capable AI model. It must persuade consumers that an AI system with access to some of the most sensitive parts of their digital lives can be trusted to act on their behalf.

The broader consumer AI market is moving in that direction. The competition is increasingly shifting from systems that can answer questions to systems that can perform tasks, navigate software, and make decisions within boundaries set by their users.