Meta launches Muse AI in US: What the new personal agent can do for users

### Meta Launches Muse AI Agent in the U.S. to Handle Tasks for Users

Meta has launched Muse in the United States, introducing a personal AI agent designed to perform tasks on behalf of users rather than simply responding to questions.

Muse can handle a range of everyday activities, including sending emails, booking travel, completing online forms and making purchases.

The service is now rolling out across iOS, Android and the Muse website in the U.S. Meta said most users will be able to access Muse for free, while paid subscription plans will offer additional capabilities.

Meta also plans to eventually bring Muse to its AI-powered glasses, expanding the technology beyond smartphones and computers.

The launch represents Meta’s broader effort to transform its AI assistant into an agent capable of taking real-world actions for users.

Unlike traditional chatbots, Muse is designed to work through multi-step tasks and continue operating even after a user closes the application.

Users can simply describe what they want accomplished, after which Muse can create a plan, coordinate the necessary resources and adjust its approach if circumstances change.

When a task reaches a stage requiring the user’s authorization, Muse can pause and request confirmation before proceeding.

For example, the agent may ask for approval before sending an email, submitting information or completing a purchase.

Meta described the system as an AI assistant that allows people to tell Muse what needs to be done while the agent handles the required actions.

The company has also introduced several security measures as it begins making the technology available to consumers.

Muse operates inside a dedicated virtual machine called Muse Secure VM, which contains the AI agent and the user’s data.

The system has its own browser, allowing Muse to interact with online services that users authorize.

A separate security agent known as Sentinel monitors Muse’s internet access and determines whether individual actions are permitted.

Meta said Muse cannot reach the internet unless the Sentinel system approves the activity, creating an additional layer of protection.

The agent also does not directly access users’ passwords or payment details. Instead, sensitive credentials are stored separately in secure storage and can be used without exposing the underlying information to Muse.

Sensitive actions require user approval, while an audit trail records actions the agent has completed and those it intends to take.

Users can control which services Muse is allowed to access and establish different permissions for individual accounts.

For example, users can allow Muse to read their emails without giving it permission to send messages on their behalf.

Muse can also remember information from previous conversations and use that context when completing future tasks.

Meta said users will have the ability to ask Muse to forget particular information and can choose not to have their interactions used to train Meta’s AI models.

For online purchases, Muse can use Stripe’s Link payment service through a wallet specifically designed for AI agents.

The wallet creates a one-time-use card for transactions, providing an additional security layer when the AI makes purchases.

Meta said Muse is the first AI agent covered by Link’s purchase protections, while support for Shop Pay and 1Password is also planned.

The company is preparing another security feature called Muse Confidential VM, which it expects to introduce later this year.

That system is designed to encrypt the entire virtual machine, including user information and conversations, with an encryption key controlled exclusively by the user.

Muse is powered by Muse Spark, which Meta describes as its most capable AI model so far and one specifically developed to handle agent-based tasks.

The U.S. launch puts Meta into the growing competition around AI agents that can independently complete tasks and interact with online services.

With Muse, Meta is moving beyond AI systems that primarily generate answers toward technology designed to perform actions and manage tasks in the real world.