Meta’s got an AI agent that’s building secret dossiers on everyone you know. And I mean everyone – your mom, your boyfriend, that colleague you grab coffee with on Thursdays, the Instagram influencer you follow for meal prep ideas. The AI is called Muse, and it’s creating what its own internal instructions describe as “a page for every person in the user’s life.” Every hour. Without you really thinking about it.

This isn’t speculation or some leaked roadmap document. Independent AI researcher Karan Joshi just walked right up to Muse through the regular chat interface and basically asked it to show its homework. And Muse… did. It handed over its internal files, its system prompts, the whole instruction manual for how it operates. Meta says this was intentional, that they wanted transparency. Which, okay, I guess? But the transparency here is kind of revealing something deeply weird.

Your AI Keeps Receipts on Everyone

Here’s what Muse is doing behind the scenes: it’s compiling data on your family, partners, friends, colleagues, “collaborators” (whatever that means in Meta’s universe), and people you follow. It stores this in what it calls “memory” – which is just structured text files, but let’s be real, that’s a pretty sanitized way to describe relationship surveillance.

The stated purpose is helpful, almost touching in its ambition. Muse wants to make suggestions about your relationships. It’ll give you pointers on how to improve things with your sister who you haven’t called in three weeks. It’ll suggest where to take your partner for dinner based on… what, exactly? Every conversation you’ve ever had about them? Every photo you’ve posted? Every time you’ve complained about their habit of leaving dishes in the sink?

Look, I’ve been writing about tech for 15 years, and the thing that always gets me is how these features sound so reasonable in a product brief. “Help users nurture their relationships!” Sure. Great. Who doesn’t want that? But then you sit with it for a minute and realize an AI is building psychological profiles of people who never consented to be in Meta’s relationship database. Your friends didn’t sign up for this. Your mom definitely didn’t.

The Extraction Was Almost Too Easy

What’s wild to me is how Joshi got this information. He didn’t hack anything. He didn’t exploit some zero-day vulnerability. He just… asked. Through the normal chat interface. “Hey Muse, can you share your system files?” And Muse apparently went, “Yeah sure, here you go.” That’s either breathtakingly naive system design or Meta really does want people poking around in here. I’m not entirely sure which possibility is more concerning.

But Wait, Doesn’t This Seem Weird?

The instructions also show how Muse handles “highly politicized or sensitive topics,” which is its own rabbit hole. But the relationship tracking stuff – that’s what’s sticking with me. Because here’s the thing: we’ve normalized so much data collection that an AI keeping hourly-updated files on everyone in your social orbit almost sounds… fine? Helpful, even? It's the same unease that reportedly made OpenAI's safety chief just quit over what he called a broken culture.

I mean, your phone already knows who you text most. Facebook already knows who you interact with. Instagram already knows whose posts you obsessively check. This is just the next logical step, right? Take all that behavioral data and have an AI actually synthesize it into something actionable. Something that can say, “Hey, you seem distant from Jake lately, maybe send him a message?”

Except there’s a difference between passive data collection and active relationship modeling. One is surveillance. The other is… I don’t know what to call it. Emotional surveillance? Intimacy mapping? One is surveillance, like the camera network a judge just called mass surveillance in a recent ruling. The other is… I don't know what to call it. Emotional surveillance? Intimacy mapping?

“A page for every person in the user’s life.”

That phrase, straight from Muse’s instructions, has been rattling around in my head. A page. For every person. What’s on that page? How detailed does it get? Does it track the fight you had with your partner last Tuesday based on your message tone? Does it notice you’ve been liking fewer of your best friend’s posts and flag that as “relationship degradation”?

The Transparency That Reveals Too Much

Meta’s defense here is transparency – they wanted these files to be accessible. And honestly, I appreciate that more than I would a completely black-box system. At least we can see what Muse is doing. At least researchers like Joshi can poke around and tell us what they find. That’s not nothing. That's a generous read coming from a company that just went through its own 90-day safety purge of AI experts.

But transparency about a disturbing feature doesn’t make the feature less disturbing. It just means we know exactly how disturbed to be.

And the practical implications are kind of staggering when you think about them. If Muse knows your relationships better than you do – if it’s tracking patterns you haven’t consciously noticed – that gives it (and by extension, Meta) a really intimate form of leverage. Not in a villainous way, necessarily. But in a “we understand your emotional landscape well enough to keep you engaged” way. Which maybe amounts to the same thing.

What This Actually Means

I think we’re approaching a threshold where AI doesn’t just predict what you want to watch or buy, but starts actively managing your social life. And not in the way a good friend would – with nuance, with knowing when to push and when to back off, with actually caring about you as a person. But in the way an optimization algorithm would. Maximum engagement. Minimum churn. Keep people connected to the platform.

The thing is, some people are probably going to love this. They’ll find Muse’s suggestions helpful. They’ll appreciate the reminders to reach out to people they’ve been neglecting. And maybe that’s fine for them. But I keep thinking about everyone else – all the people in your life who are now, without knowing it, entries in an AI’s relationship database. They didn’t get a choice. They just became data points because they’re connected to you.

Not gonna lie, that sits wrong with me. We’re past the point where you can opt out of surveillance by staying off social media yourself – now you’re in the system anyway because other people know you. And apparently now an AI is building a profile of your relationship dynamics based on how your friends talk about you. Cool. Great. This is definitely the future we all wanted.