Meta's Muse AI Is Even Creepier Than We Thought. Basically, It's A 'Digital KGB.'

Imagine downloading the AI assistant to help manage your calendar, answer emails and organize your life.

Then, that same assistant quietly builds detailed profiles of your spouse, children, friends, coworkers and practically everyone else you interact with. It remembers what they like, where they live, what you've promised them and even the arguments you've had.

Welcome to Meta's latest dystopian project.

According to a new investigation by WIRED, Muse is designed to create what amounts to a KGB dossier on the people in a user's life, continuously updating those records as it learns more.

And millions of people have already downloaded it. Muse reportedly surpassed five million downloads in just three weeks, with users connecting the assistant to personal messages, financial accounts, calendars and other sensitive information so it can carry out everyday tasks.

But independent security researcher Karan Joshi discovered something particularly disturbing when he examined the instructions governing the software.

Using Muse's ordinary chat interface, Joshi persuaded the assistant to reveal internal files describing its behavior. Other researchers have also extracted operating instructions, which Meta says were intentionally made accessible for transparency.

One instruction reportedly directs the assistant to create “a page for every person in the user’s life.”

Every person. The process runs hourly, according to researchers, gathering available information about relatives, romantic partners, friends, colleagues, collaborators and people the user follows.

These aren't necessarily simple contact entries containing a name and phone number, either. The documentation describes pages that can grow into detailed personal histories, complete with sections labeled Facts, History, The relationship, In common, Open threads and Strengthening.

One instruction calls for recording “Where they live, what they do, the threads that recur (the apartment move, the shared savings goal).”

Another suggests documenting “the trip in March, the argument that got resolved, the milestone last week.” And the assistant isn't merely keeping track of birthdays or dinner preferences.

The Lives of Others, anyone?

It is instructed to consider “how close they are, what it is built on, how they act with each other, and what it seems to need right now.” In other words, Meta has built an AI assistant that can attempt to analyze the dynamics of your personal relationships.

The system even includes a section devoted to improving those relationships, suggesting “A reason to call, a date worth remembering, something they said to circle back on, a way to be there for them that matters.”

Joshi wasn't particularly reassured.

“What it seemed like to me—from all these prompts, system skills data, and things that they’re feeding into Muse—is that they want to understand your relationships that you have with real people,” he told WIRED. “They’re trying to know you like a friend, which is honestly pretty creepy.”

That's one way to put it.

There's an enormous difference between asking a digital assistant to remind you about your anniversary and allowing software to assemble an evolving psychological and social portrait of your marriage. And remember, the people being profiled may never have downloaded Muse or agreed to participate.

Oxford University AI ethics researcher Carissa Véliz highlighted that concern.

“We are giving AI systems much more information about us than we are getting information from them,” she told WIRED. “It’s not only what we explicitly tell them, but what they can infer from us—correctly or incorrectly, both concerning for different reasons—and what they can piece together from other sources of data.”

That's a great point. Even if an AI assistant accurately records what you've told it, its conclusions about other people could be mistaken. A misunderstood text message or an incomplete account of a disagreement might become part of a profile that continues influencing future recommendations.

Meta, naturally, insists there are safeguards. The company says each Muse user has a separate virtual machine for storing personal information, preventing other users and agents from accessing that environment. Users can delete memories and revoke connections to outside services, and the assistant is designed to request approval before actions such as making purchases or sending emails.

Meta spokesperson Daniel Roberts defended the relationship-tracking functionality.

“For any agent to be useful and actually help you achieve your goals, it needs to have context about you and those you interact with,” Roberts told WIRED. “Muse gathers that based on public information and from what you’ve chosen to share, which is how it remembers the person who sent you an invoice is in fact the plumber who you previously hired to complete work in your bathroom or which flowers your spouse said they liked best.”

That's a leap in logic. An AI bot doesn't need to know everything about your social network to know what questions you've asked it. And it doesn't entirely answer the larger question of how much information an AI assistant should collect, organize and infer about people who aren't necessarily its customers.

And Meta's history makes that question particularly relevant. As RedState colleague Teri Christoph previously reported, the company reached a massive settlement with 29 states in August over allegations that Facebook and Instagram were deliberately designed to keep children engaged and that Meta improperly collected children's personal information.

The agreement could cost Meta as much as $17.1 billion over 10 years, including billions in payments and substantial changes to its platforms.

Those reforms include tighter age verification, stronger parental controls, limits on younger users' screen time and restrictions on certain features critics argue encourage unhealthy social media habits.

The settlement concerned Meta's existing social media products, not Muse. But it illustrates why the company's assurances about protecting personal information are likely to receive considerable scrutiny.

After all, Meta already operates platforms built around understanding people's interests, social connections and behavior. Muse potentially takes that knowledge much further by inviting users to connect private conversations, financial information and other material that social media platforms traditionally couldn't access directly.

Miranda Bogen, who leads the Center for Democracy and Technology's AI Governance Lab, explained the broader implications to WIRED.

“These [AI assistant] tools are actively soliciting users to plug their whole lives in—their emails, calendars, financial institutions, everything in order to be helpful assistance,” she said. “That’s dramatically more information than people might have otherwise given to some of these companies. The breadth of access to information that these tools have will lead to a ballooning of what they know about users.”

And that is the real problem. The more useful these assistants become, the more information they will gather to perform their tasks. The more information users surrender, the more comprehensive their digital profiles become.

Meta says Muse operates with user permissions and provides tools for reviewing its activities. Those protections matter, but so does understanding what you're agreeing to when you grant an AI assistant access to your digital life.

An app that remembers your friend's favorite coffee order seems helpful. An app that constructs and updates detailed records of everyone you know, including the private history of your relationships, is more like a politburo's wet dream.

Knowledge is power — and how elites might abuse our "private" data is becoming an issue of national security. Meta is gathering, centralizing, and analyzing too much sensitive data for any tech company. 

And knowing what we know about how Meta interfered in the 2020 election, there is no reason to trust the Big Tech company, or any other one, for that matter.