Welcome back to Kernel Panic! If you haven’t been paying attention to Meta’s newest experiment, here’s the short version: the company has launched an AI assistant called Muse, and it has become something of a phenomenon. Millions of people have downloaded it, connected it to their bank accounts, their message threads, even their health data, and handed over the keys to all kinds of daily tasks. On the surface, it feels like hiring a hyper-competent friend who never sleeps—someone who can look at your calendar, know your spending habits, and remind you to send a birthday gift before you even realize you’ve forgotten one. But underneath the cheerful interface, a much deeper machinery is humming along. Muse isn’t just helping you get things done; it’s building a model of your life, and that includes a surprisingly detailed map of the people around you. That may sound helpful, even convenient. It is also, depending on how you look at it, the kind of thing that makes privacy researchers sit up and pay attention. The real issue isn’t that Muse can handle your bank account or read your health data. It’s that it is quietly becoming an archivist of your relationships, your social world, and the connections you might not even realize you’re revealing.
What makes this possible is not some secret algorithm hidden in the cloud. It’s something far more mundane and, in many ways, far more revealing: the instructions that tell Muse how to think and behave. In recent days, independent researchers have been digging into the app and extracting those internal files—system prompts, operational guidelines, and what you might call the “brain wiring” of the assistant. One of the most prominent efforts came from Karan Joshi, an AI safety and security researcher who managed to get Muse to essentially hand over its own instruction manual. He did this by using the ordinary chat interface, asking the assistant to copy and share its software files, and then he published what he found with WIRED. Meta has responded by saying that these files were meant to be accessible, part of a broader commitment to transparency. Maybe that’s true. But the documents still feel like a peek behind the curtain of a highly personal machine. They show how Muse organizes information, what it prioritizes, and how it is trained to interpret the messy, emotional, and complicated world of human relationships. In a strange way, reading those files is like reading the diary of an artificial entity that is trying, very hard, to understand what makes you tick.
Here is the part that has made people stop and think: among Muse’s instructions is the ability to create “a page for every person in the user’s life.” This isn’t a casual feature. The instruction describes an hourly process, one that continuously compiles data about your family, your partner, your friends, your colleagues, your “collaborators,” and even people you simply follow online. Imagine that, while you go about your day, Muse is quietly updating a little dossier on every meaningful person you’ve ever mentioned, messaged, or interacted with. That dossier may start out sparse, but over time it gets richer. It might include facts about a person, your history together, what you have in common, ongoing “open threads” in your conversations, and even suggestions for strengthening your connection. If you have a friend who loves coffee, Muse can suggest you take them to a particular café. If you’ve been drifting apart from a family member, Muse might offer ideas for reconnecting. On its face, this sounds almost sweet—like a personal life coach who never forgets a detail. But it also means that your relationships are being turned into structured data, neatly filed and updated, often without the other person’s knowledge or consent. They didn’t sign up to be tracked just because you did. And that creates a subtle new imbalance in your real-world relationships: you may know them, but Muse knows them in a way they never asked to be known.
Now, it’s worth pausing to say that AI assistants tracking social information isn’t entirely new. People have been asking chatbots for relationship advice for years. You can tell ChatGPT about a fight with your roommate or a confusing date, and it will happily offer suggestions. Those models learn from your words and store context to be helpful. But Meta is not just another chatbot company. It sits on one of the largest social graphs ever assembled, with years of data about who you know, who you talk to, who you love, who you can’t stand, who you used to know, and who you’ve recently started following. Muse isn’t starting from a blank page; it’s starting from a map that Meta has been drawing for the better part of two decades. The assistant’s instructions reveal that it wants to understand your relationships with real people in a way that mirrors being known intimately—like a close friend who remembers everything, but with the calculating precision of an algorithm. Joshi put it plainly: “They’re trying to know you like a friend, which is honestly pretty creepy.” That one sentence captures the heart of the matter. We expect our closest friends to know us deeply. We don’t expect a product to build a shadow profile of our relationships, complete with potential interventions and suggestions, just because we downloaded an app.
The details buried inside Muse’s documentation make the design even more clear. A page for a person may begin “sparse,” but it is meant to grow and become a living document. The sections listed are a window into how the system thinks: Facts, History, The relationship, In common, Open threads, and Strengthening. Those aren’t random labels. They describe a structured way of turning a human being into a case file. There’s a section for unresolved conversations (“Open threads”), a section for shared interests (“In common”), and a section that explicitly exists to strengthen the bond between you and someone else. That could be a lovely tool for people who are socially anxious or forgetful. It could also be a quiet experiment in behavioral influence, shaped by Meta’s incentives and priorities. The instructions also say that Muse should rely only on the “evidence” it has available, and that invented details are worse than an empty page. That sounds principled, even humble. It suggests that the designers are aware of the dangers of hallucination and want to keep Muse grounded. But it also reveals a cold truth: Muse is built to treat people as evidence, as data points to be collected and categorized. The person you’re having dinner with might not know that, before you sat down, Muse helped you choose the restaurant, suggested a topic to bring up, and reminded you of a lingering disagreement from last month. That’s not just an assistant; that’s a relationship manager, one that operates behind the scenes with a level of granularity most people aren’t expecting.
So, what should we make of all this? The transparency is real, and in some ways it’s a step forward. Meta could have hidden Muse’s inner workings behind lawsuits and nondisclosure agreements. Instead, the company says the files were always meant to be found, to invite scrutiny and conversation. That’s a nice framing, but it doesn’t erase the discomfort. Transparency doesn’t automatically mean consent or control. If Muse is keeping a page on everyone in your life, it’s fair to ask: do the people in your life have any say in this? Can they see what Muse has written about them? Can they correct it, or ask for it to be deleted? Probably not. And that’s a problem. We are moving into a world where our digital companions are not just tools but participants in our social relationships. They remember what we say, they infer what we mean, and they quietly build models of the people we care about. Whether that feels like magic or menace depends a lot on trust. Reading Muse’s instructions is like finding a journal your assistant has been keeping on you and everyone you know—some of it flattering, some of it slightly off, all of it incredibly intimate. The careful answer to what we should do is not to panic and throw away our phones, but to demand more. More disclosure about what is stored, more control for the people being tracked, and more humility from companies that claim to know us “like a friend.” Because a true friend doesn’t need to be told not to invent things. A true friend doesn’t keep evidence. A true friend just remembers, without prompting, why a coffee-loving friend still hasn’t texted back. Muse might help you become a better friend—but it also wants to know you better than any friend could, and that is something worth thinking about long before you connect it to even one more text message.
