Does a software feature exist if its code has been deployed to the devices of millions of people but they can’t use it yet? Not if you work at Meta.
The company’s executives have spent the last few weeks making this semantic argument about NameTag, the in-development face-recognition system that Meta built for its smart glasses. The inevitable result is confusion, but that’s easy enough to clear up.
On June 4, WIRED reported that Meta included robust—but inactive—code for NameTag in Meta AI, the companion app for Meta Ray-Ban glasses that has been downloaded tens of millions of times. In response to our story, Andy Stone, Meta’s vice president of communications, responded in part by writing on X, “Here's a thing: Wired reports Meta didn't answer several questions about how this will work. How could we? The feature doesn't exist!” Meta removed the NameTag code from Meta AI the next day.
According to WIRED’s analysis of the Meta AI app, code for NameTag appeared in the app as early as January. In mid-February, The New York Times reported that Meta had been working on NameTag face recognition. By May, WIRED found, the core components of the NameTag code were present in the MetaAI app.
Whether the feature existed prior to Meta removing the code depends on how one defines “feature” and “exist.” Whatever one’s position, a researcher who goes by the name Buchodi reviewed the code at WIRED’s request, and was able to use the NameTag system to recognize a photograph of the face of the philosopher Michel Foucault, famed for his writing on surveillance as an instrument of power.
The claim that Meta had no way of describing how the feature works—or even would work—was further undercut last week, when Meta CTO Andrew “Boz” Bosworth did so in detail on a podcast.
On the July 8 episode of the The Most Interesting Thing in AI, host Nicholas Thompson—CEO of The Atlantic and a former WIRED editor in chief—asked who NameTag would identify during a part of the discussion labeled “What’s true and false about NameTag.” Bosworth replied: “Somebody you met in person with your glasses on who introduced themselves—or you said, ‘OK, this is David, remember this person.’ Only available to you when you’re wearing your glasses—this is a person you’ve met before. Here’s their name. They’re right in front of you … That’s what we call a NameTags feature.”
Bosworth later said of NameTag, “So, it's a thing that, um—I think would be a great feature.”
In response to inquiries from WIRED about this apparent contradiction, Meta has repeatedly stressed the conditional nature of Bosworth’s statement—that it “would be” a great feature, not that it is or will be. Spokesperson Ryan Daniels specifically highlighted the word “would,” bolding and underlining it, in an email exchange with WIRED about the apparent disconnect between Stone’s claim that NameTag doesn’t exist and Bosworth’s minutes-long description of it.
"There is no contradiction. Boz says this 'would' be a good feature, particularly to answer the blind and low-vision community members’ calls to help them identify people they’ve already met or want to remember,” Daniels tells WIRED in a statement. “While we're exploring this, it's not available to consumers today. We think it’s important that people understand this remains distinct from connecting glasses to a central database of people in the world, which is not a capability we are building."
To be clear, NameTag existed as of roughly six weeks ago. Meta had been building NameTag since early 2025—licensing face-recognition software, assembling a full detection-and-matching pipeline, and adding it to the tens of millions of phones the app runs on, where it sat until WIRED reported on it. While it was not possible for people to actually use it without specialized tools, WIRED’s analysis of the Meta AI code, as well as that of two independent experts, found a technically functional face-recognition system within the app millions of people have on their phones. That system still exists, if you take Bosworth’s discussion of it at face value.
Even so, Meta has insisted, and continues to insist, that that’s not quite true, in ways it declines to specify. Despite WIRED stating in the first full sentence of the June 4 article that the NameTag system was “unreleased” and reiterating that point repeatedly throughout, for example, Stone claimed on X that the story made this detail unclear. “This is more than shoddy reporting, it’s intellectually dishonest,” Stone wrote. “Pure advocacy-driven click bait.” Bosworth responded to Stone’s X post by calling WIRED’s report “incredibly misleading” and “absolutely dishonest.”
Meta did not answer WIRED’s questions about what its executives believed was misleading or dishonest about our NameTag reporting.
One point is clearly important to Meta: Bosworth asserted in his conversation with Thompson that NameTag will not rely on a “central database,” even though neither Thompson nor WIRED claimed or suggested it would. Instead, WIRED’s analysis of the app found, the NameTag system converts faces captured by Meta glasses and turns them into unique numerical signatures commonly known as “faceprints.” These faceprints could be compared against a face database stored on users’ devices that was populated by Meta’s servers.
The distinction between a “central” database and millions of local databases (i.e., your phone and mine) connected to Meta’s servers is not merely academic. As Bosworth mentioned during his discussion with Thompson, some state laws, such as the Biometric Information Privacy Act (BIPA) in Illinois and Texas’ Capture or Use of Biometric Identifier Act (CUBI), limit companies’ ability to publicly deploy face-recognition technology, which require people to explicitly consent to having their faces captured by face-recognition systems, among other guardrails.
Bosworth would know. In 2019, Meta abandoned its “Tag Suggestions” automatic face-recognition feature on Facebook after a $5 billion settlement with the Federal Trade Commission over privacy issues, including face recognition, and before a $650 million settlement with the state of Illinois over Tag Suggestions specifically.
By keeping face-matching on the user's phone rather than queried from a central server, Meta may be positioning itself to argue it complies with the letter of state biometric laws like BIPA and CUBI, should it deploy the feature. Whether that general design is legally defensible is contested.
In 2021, a federal judge let a BIPA class action lawsuit over Apple’s Photos app proceed, holding that a company could plausibly “possess” faceprints stored on users' own devices because possession doesn’t require exclusive control; that case was certified as a class action in June 2026 and is now awaiting a ruling on several motions.
But other courts have blessed the on-device approach: An Illinois appellate panel ruled in 2022 that Apple didn't “possess” Face ID data stored solely on users’ devices, and a federal judge dismissed a near-identical suit over Samsung's photo app in 2024 because Samsung never received or accessed the face data its own software generated.
The dividing line in these cases hasn't been where the data sits so much as who controls it—whether the feature is optional, whether users can disable it, and whether the company is capable of ever reaching the data.
When WIRED asked Meta in June whether NameTag would be opt-in and how the system retained faceprints and cropped images, it declined to answer. Asked about the question courts have treated as most important—whether the data ever leaves the device or can be reached by anyone besides the user—it would not explain why it licensed a third party's face-recognition software, when that arrangement began, or whether it continues.
Meta also did not answer WIRED’s questions this week about why the company continues to insist, even when no one is saying otherwise, that its face recognition tech does not use a central database.
Strip away the semantics, and the simple facts are these: Meta has designed and built a face-recognition system, at one point placed it on millions of phones, and appears to be seriously thinking about enabling it, and a prominent executive has sung its praises on a very public platform. Does NameTag exist? You can decide for yourself.
Additional reporting by Dell Cameron and Dhruv Mehrotra.
Facts Only
* Meta developed NameTag, a face-recognition system for smart glasses.
* Code for NameTag appeared in the Meta AI app in January.
* Core components of the code were present in the app by May.
* Meta removed the NameTag code from the app on June 5.
* A researcher identified a photograph of Michel Foucault using the NameTag system.
* Meta CTO Andrew Bosworth described NameTag on a July 8 podcast as a tool to identify people previously met in person.
* Meta VP of communications Andy Stone stated on X that the feature did not exist.
* Meta spokespersons stated the feature is not currently available to consumers.
* The system converts faces into numerical "faceprints" stored on user devices.
* Meta previously settled a $5 billion FTC case and a $650 million Illinois state case regarding face-recognition privacy.
* Illinois BIPA and Texas CUBI laws regulate the capture and use of biometric identifiers.
Executive Summary
Meta is engaged in a dispute over the existence and nature of "NameTag," a face-recognition feature for its smart glasses. While the company's communications leadership claimed the feature did not exist, technical analysis revealed functional face-recognition code had been deployed to millions of devices via the Meta AI app. Meta executives have since clarified that while the technology is being explored and described as a potential tool for identifying known acquaintances, it is not currently active for consumers.
The conflict centers on whether the feature "exists" if its code is deployed but inactive. Further tension exists regarding the technical architecture; Meta emphasizes that NameTag avoids a "central database" by storing biometric faceprints locally on devices. This distinction is critical for legal compliance with biometric privacy laws in states like Illinois and Texas. The situation remains unresolved as Meta has declined to provide specific details on whether the system is opt-in or how faceprints are retained.
Full Take
The strongest version of this narrative is that a major tech entity deployed sensitive biometric capabilities to millions of users without disclosure, then utilized semantic gymnastics to obscure the fact when discovered. This is not merely a debate over the definition of a "feature," but a conflict between technical reality (code in production) and corporate narrative (product availability).
The load-bearing pattern here is the shift between a technical "existence" and a consumer "existence." By insisting the feature "would be" great rather than "is" great, the company attempts to decouple the engineering reality from the legal and public relations liability. This allows them to test the waters of biometric deployment while maintaining plausible deniability regarding the active collection of data.
Patterns detected: ARC-0024 Ambiguity
The root cause is a clash between the agility of "dark launches" (deploying code without activating it) and the rigid requirements of biometric privacy law. This echoes a historical pattern of "move fast and break things," where the "breaking" is outsourced to the legal systems and privacy expectations of the user. The primary implication is a further erosion of informed consent; if capabilities are deployed invisibly, the user cannot meaningfully opt-out of the presence of the surveillance apparatus on their own hardware.
If this were a coordinated influence campaign, the playbook would involve "gaslighting" the observer by redefining common terms (like "exist") to make the critic appear delusional or dishonest. The actual content matches this pattern in the specific responses from Meta executives, who labeled factual reporting as "intellectually dishonest" despite the presence of the code.
Bridge Questions:
1. Does the distinction between "centralized" and "local" storage fundamentally change the privacy risk if the company controls the software that manages that local data?
2. At what point does "inactive code" on a user's device constitute a "deployed feature" in a legal or ethical sense?
3. How does the history of biometric settlements influence the way companies now architect their "hidden" features?
Sentinel — Human
The article functions as investigative journalism by synthesizing conflicting statements from Meta executives, technical analyses, and legal precedents to explore the ambiguous existence and implications of a proprietary facial recognition feature.
