Meta Sued Over Training Data for Its AI and Face-Recognition Systems

A proposed class action filed last week in federal court in Chicago claims Meta illegally used people’s Facebook and Instagram photos to train generative AI image systems—including Emu and Muse Image—and to build NameTag, an unreleased face-recognition feature tied to Meta’s smart glasses. Plaintiffs include Illinois and California parents and their children; the proposed class reaches people nationwide whose images were uploaded to Facebook or Instagram or submitted through Meta’s generative AI prompts, dating back to September 4, 2021, potentially numbering in the millions.

The complaint leans on Illinois’ Biometric Information Privacy Act and California privacy claims, seeking statutory damages for intentional, reckless, or negligent violations plus injunctive relief. It arrives after WIRED’s June reporting that NameTag code was embedded in a Meta glasses AI companion app downloaded more than 50 million times—technically functional for turning captured faces into biometric signatures compared against faceprints on-device, with a database configured to receive Meta updates, though the feature was not enabled for users.

Meta denies the suit’s merit, says it has been transparent about using people’s information to improve AI products, insists NameTag has not shipped and no final product decision is made, and repeats that it is not building a universal face database. History makes the filing weighty anyway: Meta paid $650 million in a 2020 Illinois face-recognition settlement, announced shutting down that system and deleting more than a billion faceprints in 2021, and in 2024 agreed to pay Texas $1.4 billion over separate biometric allegations.

We’re reading this as pattern recognition, not a one-off privacy scare. Meta’s own executives have called Facebook and Instagram a “data advantage” for AI. Muse Image briefly let people generate images from others’ public Instagram accounts before Meta yanked the feature and admitted it “missed the mark.” NameTag’s code in a mass-downloaded app—then removed the day after WIRED’s report, while Meta argued the feature “never existed” because consumers couldn’t turn it on—looks like shipping capability ahead of consent theater.

Useful AI glasses and better image models are legitimate product goals. Building them by treating social photos as a biometric quarry without clear notice and consent is how you recreate the last decade’s faceprint liabilities with a generative-AI coat of paint. CTO Andrew Bosworth called WIRED’s June reporting misleading, then described on a podcast a remember-people-you-met NameTag concept he thought would be great. Exploration and undeployed code are not the same as “nothing to see here.”

The lawsuit’s open uncertainty is real: the complaint itself notes Meta has not disclosed which images, if any, were used to generate biometric data. Plaintiffs will have to dig that out in discovery. That does not make the filing frivolous; it is why civil process exists when a company alone holds the training and faceprint ledger.

If this theory scales and courts treat social-photo training as biometric harvesting under BIPA-style statutes, expect tighter consent UX, narrower training sets, and higher legal reserves across the industry—not only at Meta. I’m watching whether NameTag stays shelved with full transparency if it ever returns, and whether “we’re not building a universal face database” survives discovery about how faceprints are sourced, stored, and updated.

WIRED’s September 11 story by Dhruv Mehrotra (updated with Meta’s comment) covers the Chicago federal complaint, the June NameTag reporting, Meta’s Emu training statements via chief product officer Chris Cox, Muse Image’s short-lived Instagram-based generation feature, and Meta’s prior biometric settlements. Named plaintiffs include Francisco Alvarez and his son (Illinois) and Jeremy Wahl and his 10-year-old daughter (California). Counsel for plaintiffs include Justin Boley of Wexler Boley & Elgersma.