Meta released its personal agent on September 8, it's called Muse, and it's already in second place among the most downloaded apps in the United States. 83,000 installations on the American App Store in three days. Number two. Well done. Now I'll explain why it won't be coming into my home, and why I don't think it'll be coming into yours either, at least not as long as we have a privacy law that holds up.

And first, this figure, because it teaches us something about all the apps on your phone, not just this one.

The App Store ranking does not count downloads

Apple doesn't publish its formula, but the principle is known: what it ranks is a speed. An app's progress compared with the others, over a window of a few days, in a country and in a category. A quiet week, a few tens of thousands of installations, and there you are at the top of the bill.

Look at the same day on Google Play, where the ranking works differently: Muse was sitting in 338th place in the Productivity category. Same app, two stores, two stories, and the second one is considerably less dreamy than the first. To put it in context: 83,000 in three days is roughly what ChatGPT swallowed in a single day.

A silver medal for a quiet week

An agent that needs your whole life

A chatbot answers you, an agent does things in your place. The whole difference lies in that sentence. Muse connects to your inbox, your calendar, your health apps, your smart home, your shopping sites and your payment methods. You ask it to find you a train, it compares the schedules and books it. You ask it to renegotiate a subscription, it writes the email and sends it.

It exists as an iPhone and Android app, from a browser, and through WhatsApp. Meta's smart glasses are announced as the next step. For now, none of this exists anywhere but the United States. The price, then: a free offer with weekly limits, one at 20 dollars a month, and one that goes up to 100 dollars. But you can guess that the real price isn't on the bill.

The architecture is sound, and that's not the question

On paper, I admit it's good work. Each user has their own virtual machine at Meta, in other words a personal computer hosted who knows where, on which the agent works. A second program, nicknamed Sentinel, rereads every action before it goes out. Sensitive actions, sending a message or paying, wait for your approval. And your credentials sleep in a separate vault, which lets the agent use them without ever seeing them go by.

A separate box, a doorman in front of the door, a vault for the keys. It's better than letting the program rummage directly through your phone, nobody will say otherwise.

Except that an architecture isn't judged by a diagram. And above all, a technical guarantee protects you from the agent, it doesn't protect you from anything against the company hosting the agent. The company, it has a file. And I mean every word.

Your life in a box. The doorman is there, the vault too. But META watches everything through the glass.

Meta's European record doesn't come out of nowhere

Three dates. They're public, and they're all on the bill.

On May 22, 2023, the Irish regulator, the one that monitors Meta on behalf of all of Europe, slapped it with a 1.2 billion euro fine for sending the data of its European users to the United States. It is still the biggest fine ever imposed under the European regulation on data protection. Not a slap on the wrist: 1.2 billion.

On April 23, 2025, the European Commission handed Meta 200 million for the way it charges people or gets their consent. You know this screen: either you accept advertising tracking, or you pull out your bank card. Apple got hit with 500 million on the same day, for something else: it was the first time Europe had sanctioned a company under its new law on digital markets.

And between the two, there was everything else. Meta AI, the company's assistant, waited about a year before being allowed to arrive in Europe, while the same Irish authority checked what was going to be done with the data. The version that eventually arrived doesn't even have the right to make images. Meta's model that can look at a photo still hasn't been released here: the company says European regulation is unpredictable. And in April 2025, Meta started training its models again on the public posts of its European users.

The two-button screen. The big one means accept. The small one means pay. You wonder which one is easier to aim at

Remember the month of July

This isn't ancient history, and it isn't a rumor. On July 7, Meta launched Muse Image, its image generator. And to make sure it worked well right away, the company automatically enrolled all public Instagram accounts, without sending the slightest message to the people concerned. You tagged someone in your image request, and the system went digging through their public photos.

Its own help page says so, in its lawyer's language: you won't be notified about content created with Meta's artificial intelligence features.

It took three days of uproar, from users and also from celebrity agencies in Hollywood, for Meta to remove the feature. Three days. The time it takes to realize you've made a mistake in the most visible way possible.

So yes, it was another product and probably another team. But it's the same company, the same servers, and exactly the same reasoning: put the default setting wherever it suits us, and leave the user to fend for themselves to find it again. A company that's now asking you for access to your email inbox and your bank account showed, two months ago, what it does when nobody is watching.

I'm not saying the others are clean, either. A few days ago, I told you about Anthropic backing down on the conversations it kept for 30 days: the company I use every day, and it too had pushed the setting too far. The difference is that it backed down on its own, without a fine and without a regulator behind its back. That's exactly what we ask of a company: not to be perfect, to correct itself.

The internal notes tell a different story

Internal documents published in the press don't match the launch speech. They say that an agent displayed a user's private iCloud photos on the screen while analyzing them, that pages stopped refreshing after fifteen minutes, and that errors were silently ignored.

Meta also says that its internal technical and security incidents increased by 40% in one year, and that the time spent putting out fires rose by 70%. The release of Muse, planned for April 2026, was pushed back to September, officially to strengthen security. Meta's technical director himself said that he had been disconnected from his account repeatedly.

The time spent putting out fires, while waiting for META to pump a lot of info.

Concretely, what does that change for you

Right away, nothing. Muse doesn't exist in Europe, there's nothing to uninstall and nothing to uncheck this morning.

The first thing to keep in mind is to stop reading a ranking as a sales figure. The day you see number 1 in downloads for an app, you'll know that it means it's climbing fast. Not that it's being used, and even less that it's good. That's true of Apple, Google, and everyone who cites their position in a press release.

The second, and this is the one that will decide whether you ever install it: what Muse would do beyond what you already have at home. At my place, the house runs on a little Jeedom in a corner, and my photos live on a Synology in a cupboard. I'm not going to tell you it's the same thing, my Synology doesn't book trains and it doesn't renegotiate my subscriptions. It sorts my photos, recognizes faces, finds duplicates, and none of that leaves the house. That's the real question to ask yourself when looking at Muse: what do you gain, and how much of your life do you give up to get it.

And the timeline, honestly: I haven't the slightest idea when Muse would be authorized in Europe. What I do know is that Meta can't release it here in the state it's in in the United States. Health data, payments, messaging, everything Muse swallows in the name of your convenience, that's precisely the list Europe will ask it to explain before saying yes. Given how long Meta AI took, and given that this time we're talking about a program that touches your money, I think we're talking years. Maybe never. And frankly, I won't miss it.

My opinion

What bothers me isn't the agent. I think the idea of a program that sorts my papers for me is a good one, and I think that in ten years we'll find it strange that we had to do everything ourselves.

What bothers me is the seller. You can have the most beautiful virtual machine in the world, with the best doorman and the prettiest safe, if the company that makes it has $1.2 billion in fines to its name and three days of hindsight on its own behavior in July, that doesn't reassure me. A company is judged by its incidents, not by its diagrams.

There's a second detail that makes me smile. A company that spends all its time explaining that its artificial intelligence is going to change everything should have more to show than 83,000 installations and a second place. That's the kind of figure you bring out when you have nothing else at hand.

So, to answer the question nobody asks: no, I won't install it, and if you ask me what I think, neither will you. Give Muse your bank credentials if you feel like it, you're grown-ups. I'll keep my phone for what it knows how to do very well all by itself, and as for my inbox, I'll keep opening it by hand.

What worries me isn't Muse. It's that in two years, refusing to connect an agent to your bank account will be seen as weird behavior.