Meta's New Glasses Can Log Your Face at a Party — And You'll Never Know

Imagine you're at a friend's birthday party. Someone across the room is wearing a normal-looking pair of glasses. And those glasses just logged your face — quietly, permanently — and you'll never know it happened.

That's the scenario Meta's chief technology

That's the scenario Meta's chief technology officer, Andrew Bosworth, laid out this week. If you've ever stood in a crowd, this story is about you. Bosworth described how Meta wants to put facial recognition into its consumer smart glasses. The glasses would capture the faces of people around you. Then, later, they'd whisper in your ear when they spot someone they've seen before. The catch is that the person being scanned never agreed to any of it. So the question threading through today's episode is simple. When face matching moves off the police lab and onto someone's face at a party, who's checking whether it's right?

Let's start with the tool itself. Reporters found code inside Meta's A.I. app for an unreleased feature. Its internal name is "NameTag." The idea is that your glasses read a stranger's face, store it, and recognize that person the next time you cross paths. That's a big shift. Right now, face matching is something you choose to do — you unlock your phone, you tag a photo. This turns it into something that just runs in the background, all the time, on someone else's device pointed at you.

Now, Meta says it's built this to be safer. The matching happens on the glasses themselves — no giant central database in the cloud. On paper, that sounds reasonable. But here's why some people are slow to trust it. Meta once collected and stored the faces of a billion people through Facebook. A billion. That database only got deleted after Texas and Illinois sued the company. So when Meta says "trust us, it stays on the device," a lot of people remember what happened last time.

This matters because the technology isn't perfect

And this matters because the technology isn't perfect. The Innocence Project has documented at least seven confirmed cases where facial recognition pointed at the wrong person. Six of those cases involved Black people who were wrongfully accused. Six out of seven. That's not a rounding error — that's a pattern. When trained analysts using controlled systems still get it this wrong, imagine the error rate on a pair of consumer glasses at a crowded party.

There's a number that makes this even sharper. The A.C.L.U. once ran members of Congress against a mugshot database. The system falsely matched twenty-eight lawmakers to criminal photos. Part of the problem was the confidence setting — how sure the software has to be before it calls something a match. It was running at a lower threshold than experts recommend. Now picture consumer glasses making that same guess about the person next to you on the train.

For investigators, this creates a real headache. Face matching is moving out of the forensic lab and into the hands of everyday people who aren't trained. Soon, a detective might get handed evidence that came from someone's glasses. Was that match made by a trained analyst or by a gadget? What confidence level did it use? Nobody logged it. For the rest of us, it's simpler and stranger — the video or the I.D. that puts you somewhere might come from a device you never saw.