Meta Put Face-Recognition Code on 50 Million Phones. Nobody Was Told.

Fifty million people downloaded the app. Most of them probably thought they were just getting a fun way to interact with their smart glasses. What they didn't know — what nobody told them at the time of install — was that a fully built face-recognition system came along for the ride, quietly sitting inside the app like a loaded camera with the lens cap on.

This is the story ScienceBlog.com broke open. And the more you look at the details, the more it stops being a tech story and starts being a story about a gap in the rules that protects you — a gap wide enough to drive a van through.

Here's What Was Actually Inside That App

The feature is called NameTag. According to the reporting, it wasn't a sketch or a half-finished experiment. It was three working AI models, all chained together and ready to fire.

The first model spotted a face in the camera's view. The second cropped in tight on it. The third converted that face into a string of numbers — what's called a biometric signature (think of it like a fingerprint made from the math of your face, totally unique to you) — and checked it against a stored database. Then it would flash "person recognised." Start to finish, all of that ran on your phone. Not on some distant server. On the device already in your hand.

That distinction — running on the device — matters more than it sounds. It means the capability didn't require Meta's servers to activate. Everything needed to identify a face was already sitting on 50 million phones, engineered and operational, simply waiting. This article is part of a series — start with Age Verification Api How It Works.

The timeline makes this stranger still. The code had reportedly been inside the app since at least January 2026. Meta didn't publicly acknowledge any interest in face recognition until February. And the fact that finished, functional code was already distributed across tens of millions of devices? That wasn't surfaced until June. Five months of quiet.

The Part That Actually Affects You

Here's where most tech coverage loses regular people: it gets lost in whether the code was "on" or "off." But that's the wrong question to focus on first. The right question is: did you know it was there?

When you install an app, you get a wall of legal text — the privacy policy — that almost nobody reads. (Studies consistently show most people take under a minute to "review" privacy terms that would take an average reader over four hours to actually read. You're not alone in skipping it.) Somewhere buried in that wall, maybe, is a mention of "face-related features." That is not the same thing as a plain-English alert that says: "This app contains a working face-identification system. Here is what it can do. Do you agree — yes or no?"

The Electronic Frontier Foundation documented what happened after independent researchers exposed the code: Meta stripped it out. Quickly. Within roughly a day of the public outcry, most of the face-recognition capability was gone. That's actually the tell. You don't respond that fast to something that was never a real capability. The speed of removal confirmed the code was genuinely functional, not aspirational.

"Meta's sneaky tactics in slipping the face-recognition code into its smart glasses show exactly why data privacy bills need the teeth of strong enforcement." — Kade Crockford, Director of the Technology for Liberty Program, ACLU of Massachusetts

Crockford isn't being dramatic. She's pointing at a structural problem. The current rules require companies to disclose when they collect biometric data — your face, voice, fingerprints, the body stuff that is uniquely you. But "dormant code that has not yet been activated" doesn't obviously count as collection under most existing laws. So it slips through.

The Money Trail Makes the Stakes Obvious

This is not the first time Meta has been here. The company paid a $650 million settlement over biometric privacy claims in Illinois — this was related to a facial-recognition system it later discontinued. Texas followed with a $1.4 billion settlement over the same discontinued system. As Malwarebytes noted in their coverage, smart glasses quietly normalize the idea of always-on face processing — the device is on your face, it sees what you see, and most wearers don't think of it as a camera pointed at the world the way they would with a phone. Previously in this series: Your Drivers License Is About To Become Your Ai Password.

Texas alone has collected more than $2.7 billion in biometric privacy settlements through its attorney general, according to reporting from The Next Web. That number is not a deterrent anymore — it is a signal. States are not waiting for federal law. They are building precedents case by case, and those precedents are starting to suggest that having functional code on your users' devices may eventually be treated the same as actually running it.

The Counterargument — And Why It's Not Enough

Look, nobody's saying this is simple. Meta's defense is actually coherent on its face (no pun intended): the code never ran, users never triggered it, and once it became public, the company removed most of it within 24 hours. One could argue that unactivated code in an app is no different from a software feature that's been built but not yet released. Developers do this constantly — it's called a feature flag (basically an on/off switch for a feature that exists in the code but isn't visible to users yet). Standard practice.

The problem is the scale. Feature flags in enterprise software affect a few thousand users at most. Here, finished face-recognition capability — not a rough draft, but a working three-model system — shipped to 50 million people. The ordinary privacy calculation breaks down when the install base is that large. A feature flag on a population the size of Spain is not the same animal as a feature flag in a niche B2B tool.

And there's something the quick-removal narrative buries: the code had been there for five months. The removal only happened because outside researchers found it and made noise. If they hadn't, nobody outside of Meta's engineering team would have known. That's not responsible feature development. That's disclosure by exposure.

One Concrete Thing You Can Actually Do

If you've ever looked at an app permission screen and wondered whether the list of things an app can access actually reflects everything it's built to do — that instinct is correct, and this story is why. Permissions screens show what the app is asking for right now. They don't show you what's built and waiting. Up next: That Enter Your Birthday Box Is Dead Heres What Actually Che.

Here's a useful habit: before you install any app connected to a camera-equipped wearable — glasses, earbuds with cameras, anything on your body that faces outward — search the app name plus the words "privacy researcher" or "code analysis." Real people outside the company are often looking at what's inside apps, and their findings surface before companies' press releases do. You won't always find something. But when you do, you'll find it before 50 million installs, not after.

If you've ever wondered whether a face in a photo, a profile picture, or an image from a wearable device is really who it claims to be — that's the exact problem that transparent, clearly disclosed facial comparison tools exist to solve. The difference between trustworthy face-based technology and this story is a single thing: you knowing what's happening before you say yes.

The feature is gone from the app now. The code was stripped. Meta moved fast. But here's the thing that should stay with you: the feature didn't disappear because a law required it to. It disappeared because researchers found it, the public reacted, and the company calculated that removing it was easier than defending it. That is not a consent system. That is a pressure system. And pressure only works when someone is paying attention.

The real question for the next app you install isn't "Is face recognition active?" It's "Would I even know if it was already built in?"