A class action against Meta now speaks for people who never bought anything from it. Amended claims filed this week add plaintiffs who do not own the company's AI glasses but say they were recorded when someone near them switched on the AI features, and that the resulting material could then be used to train Meta's models without their consent. Some of those bystanders are minors. The suit says "millions" of consumers may have been affected, and that people caught on camera were never given anywhere to agree to their faces, bodies, voices and personal data being recorded, reviewed, labelled and embedded in Meta's AI systems.
The original case is a consumer one. It alleges Meta misled buyers by telling them the glasses were "built with privacy in mind" and that the user "controls their data and content." The plaintiffs argue that a reasonable buyer would not expect a device marketed as protecting privacy to act as a pipeline for AI training and systematic surveillance, and that phrases like "built with privacy in mind" and "controlled by you" do not explain that everything the device sees and hears, including the most intimate moments, can be viewed by strangers abroad and used to train Meta's AI.
The case was filed earlier this year following a joint investigation by two Swedish newspapers, which found that Meta retained material captured by its AI glasses and passed it to workers for review and model training. The account came from Kenyan whistleblowers who worked at Sama, a content moderation contractor and long-standing Meta partner. Reviewers, they said, came across intimate recordings. One whistleblower told the Swedish journalists that some videos showed people using the toilet or undressing, and suggested they would not have recorded any of it had they known it would later be reviewed. Other contractors said they had seen people having sex, and documents containing personal data.
Meta told Futurism it disagrees with the allegations and will defend itself in court. The company said the glasses let people use AI hands-free and ask questions about the world around them, and that if a user addresses Meta AI, the company may review that data to improve its products and how people experience them, which is how many other companies work. It also said it filters such data, attempting to remove information that could identify a person and to take measures to protect privacy.
Read closely, Meta's answer is about the wearer. It describes a user who chooses to address an assistant and thereby accepts review of what they said. The new plaintiffs did not address anything. They walked into frame. The de-identification defence has the same shape: stripping identifiers from a clip helps with the downstream dataset, not with the part of the complaint that says a stranger in another country watched someone undress. This reads less like a rebuttal than an answer to an easier question than the one now being asked.
Scale is what makes the bystander theory more than a novelty. Meta has said it sold more than 7 million pairs of smart glasses last year, and it is not alone in the category: Google is developing smart glasses with Warby Parker, a maker of affordable eyewear, and Apple is reported to be working on adding cameras and AI to its AirPods. The consent problem those products inherit is structural. A phone camera is visible, held up, and pointed. A face-worn camera that activates on a spoken request has no equivalent gesture, and the only person with a consent screen is the one wearing it.
The interesting gap in the whole dispute is that nobody has described what bystander consent would look like in practice. Not a capture light, not a policy, but a mechanism by which a person who did not buy the device can register a refusal before their image enters a training pipeline. Meta's filtering pitch quietly concedes that the footage arrives first and gets cleaned afterwards.
If the bystander claim survives a motion to dismiss, the liability it creates does not attach to Meta's marketing copy. It attaches to the product category. Google and Apple are building toward the same premise with no more of an answer than Meta has, and the industry has been selling into that gap for millions of units.