Face Recognition iPhone: Beats Masks, Still Opens for Twins

If you have an identical twin, your iPhone's Face ID may already be unlocked for both of you. Apple's own statement admits twins who look alike could probably open each other's phones. And documented cases show twins doing exactly that, even when one of them never registered a face at all.

If you've ever glanced at your phone to open it,

If you've ever glanced at your phone to open it, this touches you. Maybe you don't have a twin. But you might have a sibling who looks a lot like you, or a kid.

It's a little unsettling to learn that something guarding your bank app, your photos, and your messages can be fooled by a relative. That reaction makes sense.

But the reason it happens is actually reassuring, once you see it. It isn't sloppy engineering. So why does a phone that shrugs off photos and masks fall for a twin?

Start with what Apple promises

Start with what Apple promises. According to Apple, the chance that a random stranger unlocks your phone is about one in a million. That's a strong number.

When your phone scans you, it doesn't save a photo. It turns your face into a mathematical map and keeps it in a protected chip called the Secure Enclave. Each time you look at the phone, it builds a new map and compares the two.

That comparison produces a score, not a yes or no. Someone has to decide how close is close enough. Set the bar too strict, and you get locked out of your own phone all the time. Set it too loose, and strangers slip in. Apple tuned that bar so strangers almost never get through.

Here's the catch

Here's the catch. That one-in-a-million figure is about random people. An identical twin isn't random. Their face geometry is nearly the same as yours.

Researchers in a study called "Double Trouble" found that face recognition methods hit over ninety-nine point eight percent accuracy on general benchmarks. Tested on identical twins, accuracy fell to eighty-eight point nine percent. That's a drop of roughly ten points, hidden behind the headline number.

For researchers, that gap is a design limit. For the rest of us, it means "one in a million" comes with fine print about who's trying.

A lot of people think, "Face ID rejected a photo

Now, a lot of people think, "Face ID rejected a photo and even a silicone mask of the owner, so it must be unbeatable." That's a reasonable conclusion. Apple's demos showed fakes failing, so people assume the lock stops everyone.

But those defenses stop forgeries. A photo isn't a living, three-dimensional face. A mask isn't skin. A twin, though, is a real face, in real depth, that really does match your map. It passes the liveness check, because it's alive. And it passes the match, because it's close enough.

Picture a bouncer who's superb at spotting strangers. He turns away nearly everyone who isn't you. But when your twin walks up with your face, he has nothing to check against. His reference card only ever showed one face, and it fits both of you.

Kids under thirteen are a related weak spot

Kids under thirteen are a related weak spot. According to Apple, the chance of a wrong match is higher for children that young. Their features haven't become distinct enough yet. Same root cause. Faces that look too alike leave the math with less to work with.

Researchers have noticed that other body checks behave differently. Studies of iris scanning show little drop between twins and non-twins. Fingerprints, by contrast, let twins in about four times more often than non-twins. So every method has its own blind spot.

The algorithm isn't broken. The input is ambiguous. When two people share the same face, no face-based lock can tell them apart.