Meta's Personal AI Muse Books Your Travel and Makes Purchases for You
Launched in the US, Muse carries out tasks for users inside connected apps. Because it needs access to your email and permission to pay, the real test is whether Meta's privacy promises hold up.
Meta unveiled Muse, a personal AI that handles everyday tasks on your behalf, on September 8 local time. You tell Muse what you want to accomplish, and it draws up a plan and carries the work out inside connected apps and its own built-in browser. It is designed to fill out forms, send emails, book travel and even make purchases. For longer jobs, it keeps working even after you close the app.
How it differs from the AI assistants you already know
Muse does not stop at producing an answer. It acts directly within external services. Where the existing Meta AI was mainly an assistant that answered questions, Muse is a separate AI that does the work for you. The model underneath it is called Muse Spark. Instead of asking how to book a trip, you hand over the booking itself.
Its focus on the everyday chores of ordinary consumers also stands out. TechCrunch contrasted Muse’s direction with rival AI companies that are building up coding and enterprise products. For users, what matters becomes not only how good an AI’s answers are, but which of their own services it can plug into and what it can get done there.
Sending email and paying: who stays in control?
Meta says users can grant and revoke Muse’s access on an app-by-app basis. For sensitive actions such as sending an email or making a purchase, Muse asks for the user’s approval. Users can also review a log of what Muse has done. Even after handing off a task, the structure leaves it to the user to decide which services Muse may reach, and whether it actually sends an email or completes a payment.
As a safeguard for data, Meta says each user gets a dedicated virtual machine. A virtual machine is a separate computing environment created in software. According to Meta’s explanation, Muse and the data from connected services live inside it, while a separate watchdog AI called Sentinel checks any actions that reach out to the internet. A Confidential VM feature, in which only the user holds the encryption key, is not available yet.
Meta says it does not share Muse conversations or virtual machine data with its advertising systems. It also says users can opt out of having their conversations used to train AI models. The question of trust is bound up with past history, too. Facebook, Meta’s predecessor, paid $5 billion in 2019 to settle US Federal Trade Commission (FTC) allegations over its handling of privacy, and accepted new restrictions.
Where you can use it, and what it costs
Muse launched in the United States as an iOS app, an Android app and on the web. You can also chat with it in WhatsApp. Meta says typical usage is free, with separate subscription tiers that raise usage limits. At launch, the subscription prices reported by TechCrunch were $20 and $100 per month. A launch timeline for South Korea and Korean-language support have not been announced.
Muse widens the range of things people hand to AI, from searching for information and getting answers to booking and buying. That also means allowing it to connect to your email account and use your payment methods. Meta says Muse cannot directly see users’ passwords or payment methods. TechCrunch raised the question of trust: whether users are willing to give Meta that role. What to watch next is how the approval steps and data protection policies Meta describes actually hold up in the live service.