LifePivo

Can Smart Glasses Identify Strangers?

A dark silhouette of a person seen from behind at a table, the crown and back of their head wrapped in a glowing pale wireframe mesh of connected dots that maps the shape of the head; at one edge the mesh breaks apart and scatters into faint drifting amber specks that fade into the empty background. A pair of heavy black-framed glasses with one softly glowing lens rests on the table in muted deep-teal light
AI-generated editorial illustration. Not a documentary image.

Explained by Maya Chen — LifePivo fictional editorial host

Meta Built the Facial-Recognition Code, Shipped It Dormant to Millions of Phones, Then Deleted It

Point a pair of Meta Ray-Bans at a stranger and ask the glasses who they are, and nothing happens. The device on sale has no facial recognition in it. It never has. That is the plain answer to the question people type most about this category of product, and it stays true through everything that follows — which is what makes the rest of the story worth telling, because at almost every point in the past two years the honest answer has been "not the glasses, but nearly."

The distance between "the glasses cannot" and "nearly" is where this sits. The hardware — a camera, a microphone, a live link to a phone and to a large language model — is a facial-recognition rig missing one piece of software. Twice now that piece has appeared: once bolted on by two students to prove a point, and once written by Meta itself and pushed, switched off, onto ordinary people's phones before being quietly pulled back out. Neither time did the product ship the feature. Both times it was one decision away.

The demo that made the question real

On 3 October 2024, two Harvard students, AnhPhu Nguyen and Caine Ardayfio, showed exactly how small that remaining gap is. Their project, which they called I-XRAY, took an ordinary pair of Ray-Ban Meta glasses and streamed the video out to Instagram. A program watched that stream, waited until it was looking at a face, and fed the face to PimEyes — a commercial face-search engine — then to data brokers such as FastPeopleSearch. Out came names, home addresses, phone numbers and relatives, in near real time, for people who had no idea any of it was happening.

The crucial detail is what was not in the loop. None of the recognition ran on the glasses. The glasses supplied a camera and a video feed; every piece of identification came from third-party tools the students strung together themselves. Nguyen and Ardayfio said the point was the warning, not the tool: "The purpose of building this tool is not for misuse, and we are not releasing it." What they had demonstrated was that the capability already existed in the open, waiting for a wearable camera pointed at strangers to make it effortless. The glasses were that camera.

That is the shape a reader should hold onto, because the coverage that followed kept collapsing it. The recurring headline — that Meta's glasses can identify you — is answering a different question from the one it appears to. The hardware could be part of such a system. The product does not do it. Those are not the same sentence, and the year that followed turned on the difference.

The plan that had a codename

In February 2026 the question stopped being hypothetical. The New York Times reported, citing an internal memo, that Meta was preparing to add facial recognition to its smart glasses, under a feature internally called Name Tag, that would let a wearer identify people in view and pull up information about them through Meta's AI assistant. One sentence from that memo, as the Times quoted it, did more damage than the plan itself: "We will launch during a dynamic political environment where many civil society groups that we would expect to attack us would have their resources focused on other concerns."

Read that for what it assumes. It does not weigh whether to build the feature; it weighs when opposition will be too distracted to fight it. That is a sentence about timing, not about consent, and it is why the report landed the way it did.

Meta's answer was to narrow the ground. Asked about the reporting, a spokesperson said the criticism was "based on speculative reporting about a product we haven't launched." That was, on its own terms, accurate — nothing had launched — and it is the exact move worth watching, because "we haven't launched it" is true right up to the moment a company does, and tells you nothing about whether it intends to.

The letter, and then seventy-five of them

Two formal objections followed. On 17 March 2026, Senators Edward Markey of Massachusetts and Ron Wyden and Jeff Merkley of Oregon wrote to Mark Zuckerberg demanding answers on how Meta would obtain consent, handle biometric data and guard against misuse, and gave the company until 6 April to respond. They framed the danger as scale rather than novelty: glasses that quietly match a face in a crowd to a name, a workplace and a social profile would, they wrote, carry "serious risks of stalking, harassment, and targeted intimidation" and erode "longstanding expectations of privacy in public spaces." Meta repeated its line — the letter rested on speculation about an unlaunched product.

Then, on 13 April, a coalition of 75 organisations organised by the ACLU of Massachusetts under the banner "Eyewear Not Spywear" asked Meta to cancel the feature outright: "The glasses pose an unacceptable threat to our right to privacy and must be cancelled." At this stage everything anyone knew about Name Tag came from a leaked memo and a codename. There was no code anyone outside Meta had seen — which is what makes what happened in June the turn in the story.

The code that was already on the phone

On 4 June 2026, WIRED reported that it had found the feature — not on the glasses, but inside the Meta AI companion app that pairs with them, the app that had been distributed to and downloaded onto millions of phones. The dormant code, labelled internally for face recognition, would have converted a face into a numerical signature — a faceprint — and matched it against a store held on the user's own device, ready to flag "person recognised" the moment the wearer looked at someone it knew.

Two things about that finding matter, and they pull in opposite directions. The first is that the machinery was not running. A security researcher who examined it put it plainly: "No part of NameTag is currently running or sending biometric data to Meta's servers today." A Meta spokesperson, Ryan Daniels, said the same: "Nothing has shipped to consumers and no final decision has been made on what to do here, if anything. If we do decide to roll something out, we will take a thoughtful approach and do so with full transparency. One decision we can be clear about — we are not building a central face database." On that reading, nobody was identified, no data left anyone's phone, and the "product we haven't launched" was exactly that.

The second thing is that it was there at all. An unactivated but complete facial-recognition system had been packaged into an app that millions of people installed to control a pair of sunglasses, with no notice that their phone now carried it. The device could not identify a stranger; the software to make it do so had already been delivered to the device's owners and was one server-side switch from working.

Meta's response to being found out was to remove it. On 5 June — one day after WIRED's report — the company shipped an app update that stripped out nearly all of it: the face-recognition technology, the code meant to trigger the "person recognised" alerts, and the models and databases built to store the signatures. Andy Stone, Meta's vice-president of communications, called the whole thing "purely exploratory" and said the company could not answer questions about it because "the feature does not exist."

Why "we deleted it" is not "we won't"

That is where the glasses actually stand, and it is a genuine outcome rather than a cosmetic one: the code is gone from the shipping app, and there is no evidence any face was ever identified through it. The Electronic Frontier Foundation, which had campaigned against the feature, called the removal a victory. It also added the sentence that keeps this from being an ending: "this quiet deletion of code does not equal a permanent change of heart."

The reason to take that caution seriously is that Meta has walked this exact line before, in the other direction. On 2 November 2021, the company shut down Facebook's face-recognition system and deleted "more than a billion people's individual facial recognition templates". Jerome Pesenti, then its head of AI, wrote that the uses of the technology "need to be weighed against growing concerns about the use of this technology as a whole," and that "limiting the use of facial recognition to a narrow set of use cases is appropriate." That was the largest deletion of faceprints in the industry's history, and it did not settle the question — five years later the same company was building the capability into a wearable and reasoning, in writing, about the political weather in which to release it. A deletion is a decision, and decisions are reversible.

This is also why the question about the glasses is not really a question about the glasses. The hardware settled nothing either way; what moves is whether Meta decides to switch the software on, and whether the law lets it. That is the same fault line running under the rest of this cluster — under what the law actually says about recording strangers in public, where the older and less forgiving statute turns out to govern the microphone rather than the camera, and under the plainer worry of whether a bystander can even tell the glasses are on. Facial recognition is the sharpest version of both: a technology that works only if the person it identifies never gets the chance to object.

What is actually true right now

Strip the year down to what can be stated without a hedge. Meta's smart glasses, as sold, cannot identify strangers; there is no facial recognition in the product. The ability to make them do so plainly exists — two students assembled it from off-the-shelf parts in 2024, and Meta itself wrote a version and delivered it, switched off, to millions of phones before removing it under scrutiny in June 2026. No one has been shown to have been identified by it. And the thing standing between the two states is not a technical barrier that might be years away. It is a choice Meta has made and unmade before, and the consent rules that decide what it may do with a face if it chooses again.

For anyone weighing whether the glasses across the café can name them, that is the honest position: not yet, not by default, and not without a decision the company has already shown it knows how to make quietly.

Sources and verification

What could not be confirmed, and was therefore left out: the exact size of the faceprint vector, which WIRED reported but which the corroborating accounts here do not state; the precise scope options Meta's engineers weighed; any figure for the app's install base beyond "millions"; and a reported mid-2026 revival of face-matching on Facebook itself, which could not be independently confirmed. This article reports what regulators, lawmakers and the companies have said. It does not tell anyone their legal rights or what to do.