Meta's 'perv glasses' lawsuit: bystander footage fed into AI training

The "perv glasses" nickname has arrived in a federal courtroom. Ryan Clarkson, founder and managing partner of Clarkson Law Firm and co-lead counsel in the litigation, filed an amended complaint on August 31 alleging that Meta turned footage captured through its AI-powered Ray-Ban glasses into training data for its AI systems. The amendment expands the case to include people who never bought or wore the glasses but were recorded by them, people Clarkson says the suit calls "bystanders," people just navigating through life while a Meta customer captured them. The case arrives just weeks after Meta agreed to pay up to $18 billion over the next decade in the largest child-safety settlement of its kind, making this the second nine-level privacy storm of the season.

The original case, filed in March in federal court in California, accused Meta of marketing the glasses as "designed for privacy, controlled by you" while sending recordings to third-party contractors who could view and label the material for AI development. The amended version pulls non-purchasers and non-users into the same claim. The complaint is blunt on the consent gap: even if a bystander notices the Glasses and agrees to be recorded, he or she cannot have consented to "the mass amalgamation of their data," including visual recordings that are "collected, stored, exploited, and visually inspected by persons overseas," because Meta made no effort to disclose those practices or to seek the bystander's consent. Clarkson says the dynamic became so visible in public that people started calling the hardware "perv glasses" - a term now doing the rounds, no longer just a shock but a label.

The case is really about the gap between product mechanics and backend behavior. The glasses combine cameras and microphones with the user's ability to ask questions about what they see, identify objects and landmarks, translate information on the go, and capture photos or video hands-free. Meta's product page highlights a capture LED that lights up during recording and users can manage or delete photos, videos, and voice interactions through software. But the plaintiffs argue that only covers a simple layer: when AI features are active, imagery and audio can be transmitted to Meta's servers for analysis and, they, ultimately used to train Meta's AI models. The marketing sells control; the complaint describes a data flow where the user's cue is control.

The charges didn't emerge in the air. In February 2026, Swedish newspapers Svenska Dagbladet and a Götheborgs-Posten published and - based on interviews with workers at Sama, a Kenya-based outsourcing company that did data annotation work for Meta - describing workers who had reviewed footage from Ray-Ban Meta glasses, including highly intimate material: people changing clothes, using a bathroom, and engaging in intimate activity, with visible financial and other personal information. The original complaint cited all the reports as evidence that Meta's privacy messaging had been handled by a human-review pipeline, with workers in Kenya who could see identifiable faces despite the promised anonymization. The amended complaint anchors those same pipelines to the bystanders: they were never the likely to see a consent ANYWHERE.

Meta disputes the allegation, and the case is still in litigation. The spokesperson told Fortune: "We disagree with these allegations and will fight them. Our glasses help you use AI hands free to answer questions about the world around you. If you use Meta AI, we may review that data to help improve our products and people's experiences - this works the same way as many other companies." The statement also said Meta filters the data to help remove identifying information and protects privacy. But the phrase "works the same way as many other companies" is a surrender on the industry: if it's true, the legal theory applies to every rival with a camera in its product.

Clarkson says the case is larger than Meta. He describes a broad push to improve AI tools by the largest corporations, and the way they do it is to find "more signal, more content, more recordings that they can feed into this machine, feeding the beast." He frames the behavior as a broader "surveillance economy," where tech companies are in an "arms race" and "barreling ahead with that as their dual focus... even if it means totally changing the way society functions and who gets the collateral damage along the way." For executives, that "feeding the beast" line is the case in a phrase: models want data, products want sunshine, and the people in the footage are increasingly a source that has to be sued.

The strategic overlap for any CEO with a camera and an AI layer. This case strips the consent gap from marketing sensors and puts it inside the training pipeline itself. The bystander is a plaintiff, the outsourcing contract in Kenya is evidence, and the train is train. Say you have a "designed for privacy" product, and the trial volunteer feed into an external annotator, the complaint resembles Meta's really defense. CFOs should be careful when they hear Google - this and the same way as the industry does is not a corporate right-of-way. Before the next launch, the board should look at the annotation supply chain, privacy copy, and the law of the camera frame: the person walking into the shot holds the word problem.