A proposed class action filed last week accuses Meta of taking biometric information from Facebook and Instagram photographs without notice or consent, and using it to build NameTag — a facial recognition system for smart glasses — and to train the generative models Emu and Muse Image. The plaintiffs are suing under Illinois and California law, covering images posted from 4 September 2021 onward, and estimate the nationwide class could run to millions of people.
The NameTag half of the case starts with reporting rather than with Meta. In June, WIRED found that NameTag's code had been quietly embedded in the AI companion app for Meta's glasses — an app that had been downloaded more than 50 million times by then. The feature was not switched on for users, but the analysis showed the system was technically capable of converting faces captured by the glasses into biometric signatures, matching them against face prints held in a database on the user's phone, and receiving updates to that database from Meta. What WIRED could not establish was where the original face prints came from.
That gap is the case. The suit argues the prints could have been generated from Facebook and Instagram photos, citing reports in which Meta employees said NameTag would be able to identify people through their Meta connections or public Instagram accounts, and a Meta patent describing the matching of faces against profile pictures and other images the company holds. The complaint concedes, in the same breath, that Meta has not disclosed which images were used to create biometric data, if any were. Only Meta knows.
The training claims are firmer, because Meta made them itself. The company has said Emu was trained on large volumes of photos and text from Facebook and Instagram, and chief product officer Chris Cox has described the platforms as a data advantage for Meta's AI systems. The plaintiffs argue that in the course of that training, Meta unlawfully collected biometric information about the people who appear in the images. Muse Image, released this summer, had already drawn criticism for letting users generate images from other people's public Instagram accounts; Meta removed the capability within days, saying it had misjudged the situation.
Meta's spokesperson called the suit baseless and said it mischaracterises how the company works, adding that Meta is open about how it uses user information to build and improve AI products. NameTag, the company says, was never released to consumers and no final decision has been made about its future; if Meta launches it, it promises to proceed carefully and disclose the details in full. The spokesperson stressed separately that Meta does not build a universal database of faces — the same thing the company told WIRED in June, when it declined to say whether NameTag would be opt-in or how it would store face prints.
Justin Boley of Wexler Boley & Elgersma, counsel for the plaintiffs, said people should not have to fear misuse of their biometric data simply because their photographs appeared on a social network. The named plaintiffs are Francisco Alvarez and his son, residents of Illinois, and Jeremy Wall and his 10-year-old daughter, residents of California. The proposed class covers residents of Illinois, California and other US states whose images were uploaded to Facebook or Instagram, or sent into Meta's generative AI systems through prompts.
Under Illinois' biometric privacy statute the plaintiffs seek $5,000 for each intentional or reckless violation, or actual damages if greater; $1,000 for each negligent violation, or actual damages if greater; and injunctive relief. The California claims carry additional compensation and further remedies.
Multiply $5,000 by a class the plaintiffs themselves describe as millions and the statutory exposure is a number nobody will ever pay. The practical range was set years ago: Meta agreed to $650 million in 2020 to settle an Illinois class action over an earlier facial recognition system, announced in November 2021 that it would shut that system down and delete more than a billion face prints, and paid Texas $1.4 billion in 2024 over unlawful collection of users' biometric data. Against a company of Meta's size, those are the real prices, and the sequence is worth reading in order: a settlement, a shutdown, a billion prints deleted — and then, four years later, working face recognition code sitting inside a glasses app with tens of millions of installs.
The defence Meta has run since June is the weakest part of its position. After WIRED published on 4 June, Meta deleted NameTag's code from the app the next day, and maintained the feature had never existed because consumers could not use it. That is a claim about product status, not about capability, and WIRED's analysis along with independent researchers' checks established the capability. CTO Andrew Bosworth called the WIRED piece deeply misleading and dishonest, then weeks later described NameTag on a podcast as a system that could recognise people the wearer had already met and asked the device to remember — and said such a feature would be useful. You cannot describe what a thing does, argue it would be good, and also hold that it never existed. Meta has continued to call NameTag an exploratory project rather than a consumer product.
The question the filing cannot reach is the one that decides the outcome. The plaintiffs are asking a court to find that Facebook and Instagram photos became face prints, and they admit they do not know whether that happened; the evidence sits with the defendant, and the case turns entirely on discovery. Everything else in the complaint — the patent, the employee statements, Cox's data advantage, even the 2004 messages in which Mark Zuckerberg reportedly spoke dismissively of the people who had trusted him with their data — is circumstantial scaffolding around a fact the plaintiffs are hoping to obtain rather than one they have.
Meta says no decision has been made about whether to ship NameTag. That is the sentence to watch. A settlement resolves what was done to photographs uploaded before September 2026; it does not touch the product decision, and on the evidence of 2020, 2021 and 2024, the cost of getting this wrong has been priced and found affordable.