Meta sued over alleged facial recognition training for smart glasses

Meta Platforms is facing a proposed nationwide class action lawsuit accusing the company of using photographs from Facebook and Instagram to extract biometric information for an unreleased facial recognition system intended for its smart glasses.

The suit further alleges that Meta has used images to train generative AI models without obtaining consent.

The 66-page complaint, filed in the U.S. District Court for the Northern District of Illinois, alleges Meta used images depicting both users and non-users to train and test facial recognition technology known internally as NameTag; to generate facial templates for the system; and to develop image-generation models including Emu and Muse Image.

The lawsuit, Alvarez et al. v. Meta Platforms, Inc., was brought by Illinois resident Francisco Alvarez, his minor child, California resident Jeremy Wahl, and Wahl’s minor daughter. It alleges violations of the Illinois Biometric Information Privacy Act (BIPA), California publicity and misappropriation law, and the California Constitution’s privacy protections.

At the center of the smart glasses allegations is NameTag, facial recognition technology developed for Meta’s Ray-Ban and Oakley glasses, as well as the Meta AI companion app.

The unreleased system’s existence became public in June, when WIRED reported finding inactive NameTag code in the Meta AI app. The code showed that the system was designed to recognize people seen through the glasses’ camera. When activated, it would convert a captured face into a biometric signature, or faceprint, and compare it with faceprints stored on the wearer’s phone.

According to the complaint, Meta went substantially beyond developing software capable of recognizing faces.

The plaintiffs allege that the company obtained photographs from Facebook and Instagram, extracted numerical representations of the faces in those photos, and created biometric identifiers that could be supplied to NameTag to identify people encountered by someone wearing the glasses.

Those numerical representations, the complaint says, can take the form of face embeddings, vectors, or templates that encode characteristics and spatial relationships among facial features.

The plaintiffs extend their biometric argument beyond NameTag to Meta’s generative AI systems. They contend that training Emu and its successor, Muse Image, on photographs containing faces caused the models to encode identity-related facial characteristics in the model parameters, latent representations, or other mathematical forms.

The proposed class action therefore raises a different issue from the question of whether Meta has actually activated facial recognition for consumers. It challenges what Meta allegedly did to build the system and the biometric database or databases required to enable identification.

Meta has disputed important aspects of that characterization.

After WIRED disclosed the NameTag code, Meta spokesperson Ryan Daniels said the company was exploring facial recognition but that “nothing has shipped to consumers” and that no final decision had been made on releasing the feature.

Daniels also said Meta was “not building a central face database.” WIRED, however, reported that its review found the system was designed to retrieve faceprints from Meta servers and store them on users’ devices.

Meta has not yet filed a substantive response to the new lawsuit.

The plaintiffs also cite a Meta patent application published in May as evidence supporting their theory about how NameTag could operate.

The application, titled “User Identity Verification without Sharing Biometric Data With Platforms,” describes smart glasses or other devices that capture facial images, generate biometric face embeddings, and compare those embeddings with biometric representations derived from profile photos, tagged photos, videos, or other images already held by a platform.

The complaint argues that the patent describes a workflow closely resembling NameTag, but a patent application alone does not establish that Meta implemented the system as described.

The lawsuit further alleges that Meta used images from Facebook and Instagram to train and test its facial recognition model. That allegation is partly based on Meta’s acknowledged use of social media content to train its AI systems.

Meta Chief Product Officer Chris Cox said in 2024 that the company’s “data advantage” included public images shared on Facebook and Instagram, and that Instagram imagery had been used to train Emu, Meta’s then-flagship text-to-image model.

The complaint says Emu was initially trained on 1.1 billion image-text pairs.

The complaint argues that those representations qualify as biometric identifiers or biometric information because they can be used to distinguish, recreate, or recognize the person depicted.

Whether representations learned and retained by a generative AI model constitute a “scan of face geometry” or biometric information under BIPA is a legal and technical proposition advanced by the plaintiffs, not an established finding.

The complaint cites research on diffusion and image-generation models to argue that these systems can retain identity-specific characteristics from their training data.

It further contends that Meta’s models may therefore contain biometric information from people who did not upload a photograph, do not have a Meta account, or are unaware that their image was included in the training data.

The plaintiffs make a similar claim about what happens after a model is trained.

According to the complaint, when a user supplies Meta AI or another image-generation feature with a photograph containing a face, the system creates an embedding, latent representation, or similar numerical representation to manipulate or reproduce that person’s appearance.

The plaintiffs argue that representation also constitutes biometric information and allege that Meta may use submitted images to further train its models.

The distinction could make the case significant beyond smart glasses. The plaintiffs are effectively asking a federal court to consider whether certain mathematical representations produced and retained by generative AI systems should be treated as biometric data when they encode sufficiently distinctive information about a person’s face.

For the Illinois plaintiffs, the lawsuit alleges that Meta violated Section 15(a) of BIPA by failing to maintain the required publicly available policy governing the retention and destruction of biometric information, and Section 15(b) by collecting or obtaining biometric information without first providing written notice of its purpose and duration and securing informed written consent.

The California claims accuse Meta of commercially exploiting people’s photographs, likenesses, and biometric information without permission, including through AI development, personalization, and identity-recognition technologies.

The proposed national class would include anyone in the U.S. whose image was uploaded to Facebook, Instagram, or to one of Meta’s generative AI models in response to a prompt.

Separate Illinois and California classes would cover residents of those states. The proposed class period begins on September 4, 2021. None of the classes have been certified yet.

The lawsuit seeks an injunction, damages, and an order requiring Meta to stop the practices alleged in the complaint. For the BIPA claims, plaintiffs seek the greater of actual damages or the statutory damages of $5,000 for an intentional or reckless violation and $1,000 for a negligent violation.

The case comes five years after Meta announced it would shut down Facebook’s longstanding facial recognition system and delete more than a billion facial templates.

Meta previously paid $650 million to settle an Illinois BIPA class action involving Facebook face tagging and, in 2024, agreed to pay Texas $1.4 billion to resolve allegations that it unlawfully captured users’ biometric data.

The new lawsuit contends that Meta’s retreat from facial recognition on Facebook did not end its use of facial biometrics. Instead, the plaintiffs allege, the company’s vast collection of social media photos became a data source for a new generation of facial recognition and generative AI systems.

Article Topics

biometric data  |  biometric identifiers  |  biometrics  |  facial recognition  |  lawsuits  |  Meta glasses  |  smart glasses  |  training