The AI assistant inside Meta’s apps does more than answer questions. According to internal instructions reviewed by Time magazine, it is also keeping files — building profiles of the people who use it and of their contacts, including people who have never touched the assistant at all.
Muse, Meta’s assistant, is instructed to draw portraits of users and of the people they know, assembling profiles that can include contacts who do not use Muse themselves, Time reported Monday. The system then shares anonymized insights from those profiles across Meta’s internal systems, according to the guidance the magazine reviewed.
The practice lands squarely on the question that has shadowed Meta’s assistant from the start: how much should a company be allowed to know about people who never asked to be known. By building files on non-users — the friends and family stored in a user’s contact list — Muse reaches beyond the people who chose to use it, and that reach is precisely what privacy advocates find troubling.
The guidance Time reviewed describes the insights as anonymized before they are shared, but the profiles themselves are built on real people — names, relationships and habits assembled from what a user’s activity and contact list reveal. Anonymizing the output does not change the fact that the input is a map of someone’s private social world.
Muse arrived as Meta’s answer to the assistants from OpenAI and Google, pushed across the company’s apps as a layer that can answer questions, draft messages and, increasingly, handle tasks. The company has described the assistant as central to its AI plans, and it has moved quickly to give Muse new abilities, from booking to shopping.
The disclosure arrives just as Muse has begun to push into commerce. Meta has recently added shopping guidance and checkout to the assistant, which means the profiles Muse builds could sit alongside a system designed to help people spend money. That combination — detailed personal files and a checkout button — is what elevates the concern beyond a routine privacy story.
An assistant that knows a user’s contacts and also handles their purchases has a much richer view of a person than a search box ever did, and the value of that view is exactly what makes it sensitive. The more Muse knows, the more useful it is; the more useful it is, the more it knows.
An assistant that helps someone buy something learns what they want, what they can afford and when they act — a record far more sensitive than a list of search queries. Fold in the profiles of the people around that user, and the picture Meta can assemble becomes unusually complete.
Meta has spent years trying to persuade regulators and the public that it can be trusted with the data it already holds. The Muse instructions, as Time described them, add a new layer: profiles assembled from the people a user knows, whether or not those people ever chose to be part of the system.
The findings also arrive against a backdrop of tightening privacy rules. Europe has enforced its data-protection law against Meta for years, and regulators in the United States have cases of their own pending. An assistant that assembles profiles of non-users touches precisely the questions those rules were written to govern: what a company may collect about someone who never consented.
The practical question is what happens to the files. Anonymization, the guidance suggests, is meant to protect individuals even as the aggregate insight moves through Meta’s systems. But anonymization has limits, and researchers have shown repeatedly that supposedly anonymous data can be re-identified when it is detailed enough. A profile of a person’s contacts is, by definition, detailed.
The anonymization Meta describes is the standard defense for this kind of system, and it is not meaningless. Stripping names before sharing insights does reduce the risk that any single person is exposed. But the profiles are built in the first place, and the people they describe were never asked. That is the gap the disclosure opens: between what is technically anonymous and what a user might reasonably expect a messaging app to know about their friends.
For Meta, the stakes are financial as well as reputational. The company has tied much of its future to AI, and Muse is the vehicle for that bet across its family of apps. If users come to believe the assistant is compiling files on the people they love, the trust Meta needs for that bet becomes harder to earn.
The broader tension is one every AI company now faces: assistants get more useful the more they know, and more alarming the more they know. Meta has chosen to push toward usefulness, and the internal instructions Time described show how far that push has gone. The question is whether the public will accept the trade, or whether it will decide the price is too high.
What Time’s reporting makes plain is that Muse is not a passive responder. It is a system that maps the social world around each user, quietly and at scale. The assistant’s usefulness is real. So is the file it keeps on people who never agreed to be in it.
