Meta’s Ray-Ban Glasses Filmed Bedrooms. Salespeople Lied.

When Swedish investigators bought a pair of Meta’s Ray-Ban smart glasses, the salesperson assured them all data stays locally on the device. That was a lie. The footage went to Meta’s servers, then to a subcontractor in Nairobi, where workers earning a few dollars an hour reviewed intimate footage of users undressing, using bathrooms, and having sex. The Meta Ray-Ban glasses privacy gap is not a data handling controversy — it is retail deception at scale, across roughly 9 million units sold.

A joint investigation by Svenska Dagbladet and Göteborgs-Posten, published on February 27, 2026, traced the full data pipeline behind the glasses — a product that moved over 7 million units in 2025 alone. English-language coverage broke on March 3, and the details are sharp enough to survive any corporate spin cycle.

The Lie Told at the Point of Sale

The Swedish journalists purchased Ray-Ban Meta glasses and were told by retail salespeople that “everything stays locally in the app and nothing is shared with Meta.” Meta’s own AI Terms of Service say the opposite: interactions may be reviewed “through automated or manual (human) review and through third-party vendors.” Both statements cannot be true. The ToS won — and the customer lost.

Here is what actually happens. When a user says “Hey Meta” and asks the glasses to analyze something, that video leaves the device, hits Meta’s cloud servers, and gets routed to Sama’s annotation workforce in Nairobi for manual labeling. Nothing about this pipeline is local. Nothing about it was disclosed at checkout.

Meta took two months to respond to the investigation. When it finally did, a spokesperson offered a single deflection: “When live AI is being used, we process that media according to the Meta AI Terms of Service and Privacy Policy.” The company’s official advice to users: don’t share information you don’t want the AIs to use. That guidance assumes users know the AIs are watching — which is exactly the piece the retail channel removed from the equation.

What Kenyan Workers Are Actually Seeing Through Your Glasses

The Swedish reporters traveled to Nairobi and interviewed Sama employees directly. Workers described reviewing footage of people using bathrooms, undressing, having sex, watching pornography, and displaying bank card details. Meta’s automatic anonymization and blurring tools — the ones supposed to strip identifiable information before human review — fail regularly in complex lighting conditions. The intimate, low-light settings where privacy matters most are precisely the ones the technology cannot handle.

One data annotator put it plainly: “We see everything — from living rooms to naked bodies.” Another worker reflected on the users themselves: “I don’t think they know, because if they knew, they wouldn’t be recording.”

The investigation surfaced a specific scenario that crystallizes the problem. A glasses wearer placed the Ray-Bans on a bedside table before their spouse entered the room and undressed — “presumably unaware she was being watched.” The spouse never activated the glasses. She never said “Hey Meta.” She never agreed to any terms of service. A worker in Nairobi saw the footage anyway.

Workers also transcribe conversations, including discussions about crimes, protests, and explicit material. The workplace culture at Sama does not encourage moral objections. As one employee described: “You are not supposed to question it. If you start asking questions, you are gone.” Sama facilities operate under heavy camera surveillance with strict personal smartphone bans — a company that surveils its own workers while those workers process surveillance footage from another continent.

Nine Million Cameras, No Meaningful Consent — Including for Bystanders

Meta sold over 7 million Ray-Ban smart glasses in 2025, more than triple the roughly 2 million sold in 2023–2024 combined. Cumulative sales sit at approximately 9 million units as of early 2026, and EssilorLuxottica is scaling production toward 10 million annual units. This is not a niche gadget problem. Every pair of glasses is a camera pointed outward — at partners, children, colleagues, and strangers who signed nothing and consented to nothing.

The glasses’ only privacy indicator is a tiny LED light on the frame that most people do not notice. At least one hobbyist charges $60 to disable it. In 2024, Harvard students demonstrated that footage from Meta’s Ray-Ban smart glasses could be fed into external facial recognition systems to identify strangers in public. The bystander is the person this system was never designed to protect — and the person it most harms.

Sama has been here before. In 2023, TIME exposed that the company paid Kenyan workers $1.32–$2 per hour to label graphic content for OpenAI while billing clients at $12.50 per hour per worker. 184 employees sued. Sama ended its content moderation work, announced a pivot to computer vision annotation — and landed exactly the work it now performs on Meta’s glasses footage. Same company. Same Nairobi workforce. Same low wages. Same exposure to disturbing content. The product changed; the exploitation pipeline did not.

Consumers have already shown they will punish AI companies for ethical failures — ChatGPT uninstalls surged 295% after OpenAI’s Pentagon deal. Meta now faces the same trust arithmetic with a product people wear on their faces.

The GDPR Problem Meta Cannot ToS Away

European Parliament lawmakers are pressing the European Commission on GDPR compliance, and the Irish Data Protection Commission — Meta’s EU regulator due to its Dublin headquarters — is monitoring the situation. The critical legal issue: Kenya lacks an EU data adequacy decision, meaning routing unblurred personal video from European users to Kenyan contractors may violate GDPR data transfer rules entirely.

Kleanthi Sardeli, a data protection lawyer at NOYB, was direct: “If this happens in Europe, both transparency and a legal basis for the processing are lacking.” Sweden’s data protection authority IMY noted that Meta cannot weaken GDPR protections when using third-country subcontractors. The EU AI Act — with most obligations kicking in August 2026 and high-risk AI rules phasing in through August 2027 — could classify the glasses’ biometric processing as “high-risk,” adding a compliance layer Meta hasn’t yet had to navigate.

The second-order effect matters for the entire wearable AI category. Every company building smart glasses — Apple, Google, Snap — now inherits a consumer trust deficit and regulators armed with a concrete horror story to cite in enforcement proceedings. The gap between promised and actual privacy protection in AI systems is no longer an abstract concern. It has a face — specifically, the faces Meta’s blurring tools failed to obscure.

The Open Question

Meta’s ToS defense technically holds if the retail misinformation came from third-party salespeople rather than Meta employees directly. But does that distinction survive a GDPR investigation — or a class action from bystanders who never had a ToS to read?

Every pair of smart glasses is a training data pipeline disguised as a fashion accessory. Meta did not sell 9 million cameras — it deployed 9 million unwitting data collectors and told buyers at the register that none of it would leave the device. The AI safety talent exodus showed what happens when the people inside these companies object. This scandal shows what happens when nobody is left to.

The Irish Data Protection Commission’s formal inquiry decision, expected in the coming months, will determine whether Meta’s consent-via-ToS model survives GDPR scrutiny — and set the template for every wearable AI product that follows.