Meta Muse Explained: What It Does, Who Can Get It, and What It Costs

Meta launched a personal AI agent called Muse on September 8, 2026, and the pitch is not "a better chatbot." In Meta's words, "it doesn't just answer questions, it actually does the work." Muse gets its own computer in Meta's cloud, its own browser, and permission to act on your accounts: sending email, filling in forms, booking travel and buying things. It is in the United States only, it is free for most uses, and Meta has not said what the paid plans cost.

Key Takeaways

• It is US-only at launch. Meta: "Muse is rolling out in the US on iOS, Android, and muse.ai, and coming soon to AI glasses." No other country is named and no date is given for one.
• You can use it inside WhatsApp, which is the detail that decides how many people meet it. Meta says talking to Muse "works just like messaging another person, in the Muse app or directly in WhatsApp."
• It is free for most things, and Meta has not priced the rest. Its exact words: "It's free for most of what people need, with subscription plans for people who want to do more." No prices are published on any Meta surface we could read.
• It can spend money, through Link built by Stripe, using a one-time-use card so your real details stay hidden. Meta says Muse is the first AI agent covered by Link's purchase protections.
• Nothing it does reaches the internet unopposed. A separate Sentinel agent runs alongside it: "Nothing Muse does reaches the internet unless the Sentinel approves it."
• It keeps working when you close the app, and comes back "when something changes or when it needs approval, like before it sends an email or makes a purchase."
• Training is opt-out, not opt-in. Meta says "people can also opt out of their interactions being used to train Meta's AI models."
• Meta says it will still get things wrong: "Muse can and will still make mistakes."
• Full encryption is not here yet. Meta says Muse Confidential VM, where only you hold the key, arrives "later this year."

What Muse Actually Does

It carries out multi-step tasks on your behalf rather than telling you how to do them. Meta's examples are ordinary errands, not demos: selling a car for more than you would have got, lowering a bill, adjusting a training plan when your week changes, turning a recipe reel you saved on Instagram into a grocery list and remembering your friends' dietary restrictions before it sends the dinner invites.

The mechanics behind that are what separate it from a chatbot. Meta says Muse "can open a browser, fill out forms, and negotiate on their behalf," and that for longer jobs "Muse keeps working after people close the app, and comes back when something changes or when it needs approval." If you want the general version of this idea, our guide to what an AI agent actually is covers how the same pattern shows up in ChatGPT, Gemini and Claude.

The model underneath is Muse Spark, which Meta calls "Meta's most capable model to date, built for real-world agentic work."

Where You Can Get It

The United States, on four surfaces, with a fifth coming. Everywhere else, nothing has been announced.

What It Costs

Meta has published one sentence about money and no numbers: "It's free for most of what people need, with subscription plans for people who want to do more."

That is genuinely all Meta says. There is no pricing page linked from the announcement, and muse.ai redirected to a sign-in endpoint that returned an error rather than a page when we tried to read it on September 9, 2026. Reports elsewhere name monthly figures; none of them trace to a Meta surface we could open, so this page does not repeat them. If you see a price quoted for Muse, check whether the source is Meta.

What you can act on today: the free tier exists and Meta says it covers most ordinary use, so there is no reason to reach for a card before you have tried it. If you already pay for ChatGPT Plus or Google AI Pro, nothing about this launch tells you to switch, because you cannot yet compare what you would be buying.

The Money and Email Question

This is the part worth reading slowly, because Muse is asking for more access than a chatbot ever has. Meta's own descriptions of the limits are specific, which is a good sign, and they are worth knowing before you connect anything.

Two of those rows deserve emphasis because they cut the other way. Training is opt-out. Meta offers the control but does not say it is off by default, so if you care, go and find that setting rather than assuming. And end-to-end encryption is a promise, not a feature: until Muse Confidential VM ships, the VM holding your data and conversations is Meta's to reach.

How Much Should You Trust It

Meta's own answer is the honest one: "Muse can and will still make mistakes, but we expect they'll be much less frequent and cause much less damage due to the safety systems we've built in." That is a fair description of where agents are in September 2026, and it is more candid than most launch posts manage.

The structural protection is the Sentinel: a separate agent running on the same machine, "kept apart from Muse at the system level," that every outbound action has to pass and that Muse cannot override. Credentials go into secure storage rather than to the agent, "including passwords a person types into the browser themselves."

Meta has also put a price on finding holes in it. Its security post opens the Muse bug bounty to anyone and "awards up to $300,000 for valid reports, including up to $130,000 for successful prompt injection attempts that affect one user." A six-figure bounty specifically for prompt injection is a company telling you which attack it expects, and prompt injection is exactly the risk that comes with letting software browse the web on your behalf. It is the same warning OpenAI attaches to ChatGPT's Computer History.

Should You Use It

Try it if you are in the US, you already use WhatsApp, and there is a specific errand you keep putting off. The free tier costs nothing and the WhatsApp route means no new app. Start with something reversible.

Connect the minimum. Give it read access before send access on email, and add services one at a time. Meta lets you set this per app, so use that rather than accepting a default.

Turn off training first if that matters to you. Meta says the control exists; it does not say it is on your side by default.

Wait if you want the encryption. Muse Confidential VM is the version where Meta genuinely cannot see your data, and it is not out. "Later this year" is a save-the-date.

Ignore it entirely if you are outside the United States. Meta has named no other country and no timeline, so there is nothing here for you to decide yet.

Sources: Meta's announcement, Introducing Muse: The World's First Personal AI Agent Built for Everyone, published September 8, 2026, and its security and safety post. Every quotation on this page comes from one of those two. Meta publishes no prices for the paid plans, and muse.ai did not return a readable page on September 9, 2026, so no price appears here.