Meta once walked away from facial recognition on its smart glasses. Now the technology sits dormant inside the companion app millions already use. The discovery, made public this week, changes how privacy advocates view the company’s ambitions. And it raises fresh questions about what happens when cameras on your face gain the power to name strangers in real time.
The Ray-Ban Meta glasses look ordinary. That ordinariness is the point. They record video and audio with only a small indicator light. Sales topped seven million pairs last year. The New York Times first reported in February that Meta planned to add a feature called Name Tag. Wearers could identify people in view and pull details through the built-in AI assistant. Internal documents showed the company hoped a turbulent political climate would blunt criticism. One 2025 memo stated the launch would occur “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.”
But the story grew sharper this month. On June 4 a researcher examined version 273.0.0.21 of the Android companion app Stella. He found a complete on-device facial recognition pipeline. Three models. A local database. A cosine-similarity vector index. The full stack. Buchodi.com detailed how the system detects a face, generates a 2048-dimensional embedding, searches stored profiles, and fires a notification that reads “Person Recognized.” Unrecognized faces are written to disk for later processing. The code is not active for ordinary users. Yet it ships by design.
Short. Direct. The machinery is already in users’ hands.
Meta considered the feature years ago. In 2021 the company pulled back over technical and ethical hurdles. Renewed interest followed strong sales and shifting political winds. The glasses already feed photos, videos and voice recordings into AI training by default. Privacy settings changes in 2025 made opt-outs harder. Critics say the combination creates an always-on biometric collector that bystanders cannot avoid.
Advocates reacted fast. In April more than 70 organizations signed an open letter to Mark Zuckerberg. The ACLU, EPIC, Fight for the Future and groups focused on domestic violence, LGBTQ rights and immigrant protections demanded Meta abandon the plan. “People should be able to move through their daily lives without fear that stalkers, scammers, abusers, federal agents, and activists across the political spectrum are silently and invisibly verifying their identities,” the letter read. They warned the technology would put survivors at risk, allow identification at protests or clinics, and erase public anonymity. WIRED covered the coalition’s call. Meta responded that competitors offer similar products but the company would take a thoughtful approach if it moved forward.
EPIC sent letters to the Federal Trade Commission and state attorneys general urging them to block rollout. The group cited grave risks to safety and civil liberties. Illinois and Texas previously extracted more than $2 billion from Meta over past biometric practices. Those settlements loom as regulators consider new complaints.
The Technical Reality Hides in Plain Sight
The Stella app’s code reveals an engineering commitment that goes beyond speculation. Models based on open-source architectures handle detection, alignment and embedding generation. The local database uses SQLite with a vector extension tuned for 2048-float vectors. Matches trigger high-priority notifications with deep links to a profile screen not yet visible in the stock build. Everything persists across reboots. This is not a prototype. It is production-grade infrastructure waiting for activation.
Yet activation carries legal and social weight. Real-time identification in public spaces collides with consent requirements under laws like Illinois’ Biometric Information Privacy Act. Children photographed by parents wearing the glasses could feed training data without clear safeguards. Pickup artists have already used the current camera-only version to record encounters and share them online, leading to harassment. Adding names makes the harm more precise and immediate.
But sales keep climbing. EssilorLuxottica, Meta’s partner, reported tripled revenue from the AI glasses line. Consumers like the convenience. They share what the glasses see, ask the AI for information, and enjoy hands-free calls. The privacy cost stays abstract until someone points the device at you.
Meta has shut down facial recognition systems before. In 2021 it retired the feature on Facebook amid backlash and regulatory pressure. Company statements still emphasize caution. Yet the dormant code, the internal timing memo, and the relaxed privacy reviews in early 2025 suggest momentum in another direction.
Industry watchers note the shift. Smart glasses moved from novelty to mainstream faster than many predicted. The next models may include displays or more advanced sensors. Each step tightens the integration between camera, AI and personal data. Once Name Tag ships, even in limited form that only matches your contacts, the precedent is set. Expansion becomes easier.
Regulators face a choice. They can investigate now, while the feature remains gated. Or they can wait until millions of devices carry live biometric capability. Past patterns show enforcement often lags deployment. The coalition letter and EPIC filings aim to shorten that gap.
So the glasses keep selling. The code keeps shipping. And the debate grows louder. Privacy in public spaces, once taken for granted, now depends on corporate restraint and timely oversight. Meta’s next move will test both.
WebProNews is an iEntry Publication