Meta has unveiled its new personal AI agent, Muse, which it says can actively perform tasks on behalf of its users. Announced by CEO Mark Zuckerberg via Instagram on Tuesday, the agent can send emails, book travel, fill out web forms, and negotiate on a user’s behalf. It runs on Muse Spark, which Meta describes as its most capable model built for agentic work.
The agent is rolling out in the U.S. as a stand-alone iOS and Android app. It is also available at muse.ai and works inside WhatsApp. It is free for most uses, with subscription plans for people who want the agent to handle longer or more complicated tasks. Meta says its AI glasses will soon support Muse as well.
Muse accesses a user’s personal information, web credentials, and payment information to “work on a person’s behalf across the apps they use daily, learning from conversations, reflecting on what matters to them, and getting sharper along the way,” the company says.
For example, it can turn a saved Instagram recipe reel into a grocery list, suggest a dinner-party menu, and recall guests’ dietary restrictions before sending invites. For longer, more complex agentic tasks—like negotiating a higher selling price for a car or lowering a bill—Meta says Muse keeps working in the background after the user closes the app, and returns when something changes or it needs approval. For purchases, Muse can check out using an agent wallet called Link that’s built by Stripe.
Security and trust
The biggest barrier to users handing over their personal affairs to a Meta AI agent may be trust. Meta’s core advertising business is built on harvesting the personal data of its social app users, and the company has shown a disregard for the mental health of young users. The more agency and autonomy an AI possesses, the greater the potential for privacy or security breaches.
“[T]he consequences of agents getting things wrong aren’t just a bad recommendation or a creepy ad,” Miranda Bogen, chief technologist at the Center for Democracy and Technology, said in a statement Tuesday. “With minimal input from users, these systems could take actions that damage someone’s reputation, leak sensitive medical information, or even deplete bank accounts.”
Meta says Muse does not share a user’s conversations or virtual machine data with Meta’s ad systems, and users can opt out of having their interactions used to train Meta’s AI models. Users can choose which apps Muse connects to and how much access it gets to each. For email, Meta says users choose whether Muse can only read messages or also send them on their behalf. Access can be changed or a service disconnected at any time. Users can also tell Muse to forget specific things it has learned.
The security-and-trust focus extends to the architecture itself. Each Muse user gets a dedicated cloud virtual machine, called Muse Secure VM, with its own browser, which houses the agent and stores the user’s data and credentials. Meta says it will introduce Muse Confidential VM later this year, encrypting the entire virtual machine, including user data and conversations with Muse. The user, rather than Meta, will hold the decryption key.
A separate “Sentinel” agent runs on the same virtual machine. Its job is to make sure no Muse action or output reaches the internet unless it meets certain privacy thresholds, and to ask the user for permission when needed.
Meta says Muse has no visibility into passwords or payment methods. Credentials go into secure storage that Muse can use without seeing them, including passwords the user types into the browser. Muse checks with the user before sensitive actions such as sending an email or making a purchase, and shows a complete audit trail of what it has done and plans to do. The Link wallet, Meta says, generates a onetime-use card so that real card details stay hidden.
Context
Over the past year and a half, Meta has spent copious amounts of money on the AI research talent and computing power needed to develop frontier models that can compete with top artificial intelligence labs like Anthropic and OpenAI. Some analysts have said the main motivation for developing such models was to supply the intelligence for highly autonomous personal AI agents that could be distributed through Meta’s dominant social media platforms.
“[P]ersonal superintelligence will be one of the most transformative technologies of a lifetime,” the company says in a press release. “Muse is a first step: an agent that takes on more of the work so people can focus on what matters to them.”