Overview
Ray-Ban Meta smart glasses are ordinary-looking Wayfarers and Skylers with two microphones, an open-ear speaker, and a camera above the left hinge. Meta builds the software and the AI; EssilorLuxottica, which owns Ray-Ban and Oakley, builds and sells the frames. The second generation shipped in 2025, and a display model with a small in-lens screen followed.
The scale changed fast. EssilorLuxottica reported in February 2026 that AI glasses sales more than tripled year on year, and industry reporting put combined Ray-Ban and Oakley smart glasses volume at roughly 7 million units in 2025, against about 2 million cumulative from the October 2023 launch through early 2025. Bloomberg reported in January 2026 that Meta and EssilorLuxottica were discussing raising annual capacity toward 20 million units or beyond by the end of the year. Whatever you think of the category, it is no longer a curiosity. It is the first consumer wearable camera to sell in numbers that make it normal to be filmed by a stranger’s glasses.
That is the honest frame for rating them. Most products in this Index take data from the person who chose to use them. This one takes data from everyone in front of the wearer, and those people were never asked.
What It Knows About You
The glasses have no display of their own on the base model and no independent connection; they pair to a phone running Meta’s companion app and a Meta account. Photos and video land on the phone, and Meta’s optional cloud media setting keeps captures on Meta’s servers for 30 days.
The consequential change came on 29 April 2025. Meta revised the privacy policy so that Meta AI with camera is on by default, and removed the setting that let owners decline cloud storage of their voice recordings. Meta’s own policy language, as reported by TechCrunch and others at the time, states that the option to disable voice recording storage is no longer available, and that transcripts and audio can be retained for up to a year to improve Meta’s products, including model training. The remaining control is manual deletion, one recording at a time, from the companion app. There is no bulk purge and no export of your AI history.
For bystanders there is no setting at all. Their faces, their voices, and whatever was on the table in front of them go wherever the wearer’s footage goes. The only notice they get is a white LED near the camera, which cannot be switched off on shipping hardware and which Meta hardened in July 2026 so the camera disables itself if the light is detected as tampered with or destroyed.
The Real Risks
Human review of intimate footage is now the subject of a class action. Swedish outlets Svenska Dagbladet and Göteborgs-Posten reported in March 2026 that contractors in Kenya labelling training data for Meta’s glasses were seeing footage users had shared with the AI, including people undressing, using bathrooms, and handling financial documents. One worker was quoted saying they see everything. On 4 March 2026 plaintiffs Gina Bartone and Mateo Canu filed a proposed class action against Meta Platforms and Luxottica of America, alleging the companies failed to disclose that video captured by the glasses reaches servers and then a Kenyan subcontractor for manual labelling. The suit is unresolved and the allegations are allegations. Meta’s public response has been that privacy was built into the camera from the ground up and that the LED cannot be turned off, which does not address who watches the footage afterwards.
The glasses are already a working harassment tool. Wired reported in March 2026 that men were using them to film unsolicited street approaches to women and posting the results, with the women often finding out only when the video surfaced. CBS News interviewed one of them. In late July 2026 Instagram head Adam Mosseri said the platform was fighting the practice, and enforcement so far amounts to a small number of deactivated accounts. A platform policy is not a hardware fix.
Facial recognition is the capability everyone is circling. In October 2024 two Harvard students, AnhPhu Nguyen and Caine Ardayfio, wired a pair to PimEyes and public records to build I-XRAY, which identified strangers on campus and pulled names, addresses and phone numbers in near real time. They declined to release it. In June 2026 Wired reported finding an unreleased system called NameTag inside Meta’s companion app, designed to turn faces captured by the glasses into biometric signatures matched against a local database, with unrecognised faces cropped and stored for later. Meta stripped most of it out within a day of the report. Meta has separately said it is exploring name recall as a feature, and prototypes codenamed Aperol and Bellini are reported to run a background “super sensing” mode for hours rather than the roughly 30 minutes Live AI manages today. Zuckerberg has reportedly questioned whether the LED needs to stay lit in that mode. None of this is shipping, and it should not be scored as if it were, but it is the announced direction.
European regulators are debating a ban, not a fine. Thomas Fuchs, Hamburg’s data protection commissioner, told German broadcaster ARD in mid-2026 that smart glasses function as disguised cameras and could fall under German law prohibiting recording devices concealed in everyday objects. His office tested the recording LED and found it hard to see reliably in daylight. France’s CNIL has described the risk as surveillance that is close to invisible and constant, and the European Data Protection Board has taken up the question of whether bystanders have any meaningful way to object. That is a live regulatory posture in a market of 450 million people.
On bias, the record is thinner and worth stating plainly. There is no substantial published audit of output skew in the glasses themselves. What exists is upstream: the TrueSkin benchmark published in 2025 measured multimodal models on skin tone recognition and reported Llama 3.2, from the model family behind Meta AI, at roughly 44% accuracy with a consistent pull toward classifying darker and intermediate tones as lighter. The score here reflects that plus the obvious distributional question of who ends up in front of these cameras most often.
Alternatives
- Camera-free audio glasses. Amazon’s Echo Frames carry microphones and bone-conduction speakers with no camera at all, which removes the bystander problem entirely. Solos AirGo models do the same at around 35 grams with prescription support, routing voice queries to a third-party assistant.
- Assistive-first hardware. For blind and low-vision users, who have the strongest genuine case for a wearable camera, the Envision Ally build on Solos frames is purpose-designed for the task rather than being a social camera with accessibility bolted on. Judge it on its own terms.
- Your phone. Everything the glasses do well, a phone does with the significant advantage that raising it is a visible act. The friction is the feature.
- Buy them, then turn things off. If you want the hardware, open the companion app on day one, decline cloud media storage, and set a recurring reminder to delete stored voice recordings, because Meta will not stop keeping them and there is no bulk option. Treat any indoor or private setting as somewhere the camera should be off.
- Plain Ray-Bans. They cost less, last longer, and nobody backs away from you at a party.
Our Verdict
C, use with caution, driven almost entirely by one axis. The privacy score of 9 is the highest this Index has given, and it is earned by things that already happened: an opt-out removed in April 2025, a year-long retention window for voice with only per-item deletion, an unresolved class action over contractors watching intimate footage, and a European regulator openly discussing prohibition. The grade lands at C rather than lower because the other four axes are genuinely unremarkable. These glasses do not displace paid work, they are not a subscription trap, and the hardware keeps working as sunglasses if you walk away. Averaging across five dimensions is the method, and this is what the method says.
So read the C as “the specific danger is concentrated, not diffuse.” If your concern is your own workflow or your own money, this product is fine. If your concern is what you are doing to the people around you, it is one of the worst things in this Index, and no amount of careful settings management on your part fixes that, because the exposure is theirs.
What would move this. Restoring an opt-out for voice storage, adding bulk deletion, and publishing who reviews footage and under what terms would take privacy down meaningfully. Shipping face recognition, or letting the recording LED go dark in a background sensing mode, would push it to F and the verdict to high-risk. A German or French prohibition would change the entry outright. As of August 2026, the shipping product is a camera on a stranger’s face with a light you may not notice.