Meta sued over training data for its AI and face-recognition systems
WIRED reports a proposed class action accusing Meta of illegally harvesting people’s Facebook and Instagram photos to train generative image systems and to support an unreleased NameTag face-recognition feature tied to Meta glasses. Plaintiffs include Illinois and California residents; the proposed class stretches nationally back to September 4, 2021. Meta says the suit is without merit and that it is not building a universal face database.
Quick Take
- Fact (complaint / WIRED): Suit alleges biometric harvesting from social photos for AI image models and for NameTag-style face matching, including faceprints that could update from Meta infrastructure onto a phone-side database.
- Fact (prior WIRED reporting cited): NameTag-related code was found embedded in a Meta glasses AI companion app with tens of millions of downloads, even though the consumer feature was not enabled.
- Fact (Meta statement): Company calls the lawsuit meritless, says it has been transparent about AI training uses, insists NameTag has not shipped, and denies a central/universal face database.
- Judgment: Separate shipped product harm from unreleased feature risk and from training-data theories that courts still have to test. Policy pressure is real either way.
What the complaint is actually aiming at
Two threads sit in one filing. First, generative image training: Meta has publicly talked about training Emu-class systems on large volumes of Facebook and Instagram media, with executives describing social graph data as an advantage. Plaintiffs argue that pipeline swept up biometric identifiers without the notice and consent regimes some state laws demand.
Second, face recognition: WIRED’s earlier glasses reporting described a system designed to turn faces captured by wearables into biometric signatures and compare them to faceprints stored on-device, with updates from Meta. The complaint links that design to profile photos and social connections. Meta answers that nothing has shipped to consumers and no final product decision is locked.
Why builders should care before a verdict
Even pre-judgment, this is a procurement and product-design story. If you fine-tune on social scrapes, mirror “tag people in the frame” features, or ship glasses/agents that identify bystanders, you inherit the same consent questions. Illinois biometrics law has already shaped big settlements industry-wide. California plaintiffs widen the map. “We trained on public posts” is not a complete risk memo.
For Meta’s own Muse Image experiments that briefly let people generate from others’ public Instagram accounts, the company already had to reverse course after saying it missed the mark. That history is why NameTag optics are radioactive even while disabled.
What to watch next
Watch whether discovery forces clearer statements on which photo corpora fed faceprint systems, whether NameTag stays vapor, and whether training disclosures get more granular for EU/US users. If you are a competitor shipping face features, assume screenshots of Meta’s statement will be quoted at you in your own reviews.
Bandwagon Check
Fairly hyped as a serious biometrics-and-training class action against Meta’s AI and glasses stack; overhyped if treated as proof NameTag already ran at consumer scale or that every generative training use is already illegal everywhere. Fact: filed class theories, glasses code reporting, Meta denial of a universal face database, no consumer NameTag ship. Company claim: transparent AI improvement, thoughtful future approach if anything launches. Judgment: track the case; do not ship lookalike face features on vibes.
Consent UX is now a model feature
Product managers still ship “use my camera / use my photos” toggles as afterthoughts. This filing is a reminder that biometrics claims travel with screenshots of those toggles. If your roadmap includes identifying people in frames, generating likenesses, or training on social corpora, budget legal review before growth review. Meta’s denial of a universal face database is a clear public line competitors will be asked to match in plain language.
Sources
By Sean Smith · AI Bandwagon