AI data privacy just got a lot less abstract. This week, Wired reported that security researchers extracted the instructions that tell Meta’s Muse app how to gather and use personal information, and Social Media Today walked through the findings on October 4. Those instructions read like a job posting for a very thorough personal assistant with a very long memory. Worth a slow read, even if you never plan to open the app.
Some context first. Muse is Meta’s push into personal AI agents, and it’s catching on fast. Social Media Today reported last week that the app passed five million downloads and roughly three million weekly users. The pitch is guidance on nearly every part of daily life, including health, money, and relationships. To do that job, an assistant has to know a lot about you. It also has to know about the people around you.
According to Wired’s reporting, one instruction tells Muse to build a page for every person in the user’s life. The process reportedly runs hourly. It pulls together details on family members, partners, friends, colleagues, collaborators, and people the user follows. Muse stores what it learns in structured text files that serve as its memory.
The details it’s told to track are personal. Where someone lives and what they do for work. Storylines that keep coming up in conversation, like a big move or a savings goal two people share. Dates that matter, such as birthdays and anniversaries. Backstory too, including a recent trip or an argument that finally got settled.
If the goal is a relationship coach, the design makes a certain kind of sense. Meta’s CEO has pitched AI as something that can help people be better friends, parents, and partners, and you can’t coach someone on their friendships without knowing who the friends are. Still, read that list again slowly. It’s a running file on your mother, your business partner, and your college roommate, refreshed every hour, and the people described in those files weren’t part of the conversation about whether to collect it.
Keep these two things separate, because they carry different weight. The reported part is that the instructions appear to direct Muse to build and refresh profiles on the people in a user’s life. The worry comes from Social Media Today’s analysis, which notes that Meta earns roughly 97% of its revenue from advertising and suggests some of these insights could inform ad targeting. The piece also raises other possibilities: keeping users hooked on Meta’s apps, shaping political content, a leak. It points to Meta’s history with user data as a reason to be cautious.
Those are the concerns the writer raises. The reporting does not confirm that any of them is happening. We think the distinction matters, because it’s easy for a story like this to slide from “here is what the instructions say” to “here is what Meta is doing,” and those are two very different claims.
We buy media on Meta. Our clients’ players, subscribers, and customers spend time on these platforms. And most teams in this industry use AI tools somewhere in the workday now, ours included, for research, drafting, and analysis, often with client material close at hand.
In gaming and real-money gaming, a person’s data can be unusually sensitive. Where they are, how old they are, how much they spend, and how often. Habitually pasting things into a chat box without thinking is a risk in any business. Here it’s a bigger one, which is why AI data privacy belongs in the same conversation as creative and media strategy.
These AI data privacy rules apply to any AI tool, Muse or otherwise. None of this is legal advice, and a tool’s settings will change faster than any blog post can keep up, so treat the list as habits rather than a checklist for one product.
Rules are easy to nod along with and easy to forget by Thursday, so here’s a concrete picture. Say a media buyer wants help summarizing a weekly performance report. The quick version is pasting the whole export into a chat: the client’s name, campaign names, spend by day, install counts by region, maybe a few player-level rows that came along for the ride. It takes ten seconds. It also hands a third party a snapshot of a client’s business that the client never approved sharing.
The safer version takes about two minutes. Swap the client name for “Client A.” Delete any player-level rows. Round the spend to the nearest thousand. Strip campaign names that give away unreleased titles or promotions. Then ask the same question. The answer you get back is usually just as useful, because the analysis never depended on the client’s name in the first place. Good AI data privacy habits mostly come down to a small pause before pasting and asking what the tool really needs to do the job.
The same logic works for personal use. Asking for advice on a hard conversation with a coworker works fine as “a coworker who keeps missing deadlines.” It doesn’t need their name, their title, or the details of their last review.
If a team handles client or player data, rules that live in people’s heads won’t hold up. A one-page AI data privacy policy covers most of it. Which tools are approved for which kind of work. What never goes into any AI tool. How to anonymize before pasting: swap client names for placeholders, strip player IDs, round the numbers. And who to ask when a case is unclear.
Read your client contracts and NDAs while you’re at it. Some restrict what you can share with outside tools, and some clients have strong opinions even when the contract is silent. Bring in your own counsel for anything with legal weight.
It also helps to ask a few questions of any AI vendor before the team adopts the tool. Where is the data stored, and for how long? Is it used to train models? Can the vendor’s employees read conversations? How do you delete everything, and does deletion reach backups? What happens to the data if the company gets acquired? A vendor that can’t answer those clearly has told you something already.
Meta has a long record of answering privacy worries with usefulness, and so far people keep downloading. Maybe this is another round of that. Maybe the Wired findings make enough noise to change behavior. We don’t know, and whoever claims certainty is guessing.
What seems safe to predict is that the next assistant will be more capable and more curious about the people in your life than this one. Settings screens will change, and policies will get rewritten. The habit of asking what’s being collected, and by whom, travels from tool to tool. AI data privacy is a skill more than a setting, and it’s worth building now while the stakes are still easy to see. The teams that treat it that way will have less to untangle later, whichever assistant ends up in everyone’s pocket.
If your team is figuring out how AI fits into creative and UA without risking client or player data, we’re happy to talk it through. Reach out and let’s compare notes.
