NameTag is the name Meta gave to an unreleased face-recognition feature for its smart glasses, and since last week it is also the name at the centre of a proposed class action. On Friday 11 September 2026 WIRED reported that Meta had been sued over the training data behind that feature and behind its image-generation models. The complaint, Alvarez v. Meta Platforms, Inc., was filed in the Northern District of Illinois on 4 September, a week before the story ran.
It says Meta pulled faces out of Facebook and Instagram photos without notice or consent, both to build the faceprints NameTag would match against and to train the Emu and Muse Image models.
We read the 66-page complaint rather than the coverage of it, along with the five WIRED stories that preceded it since April, the coalition letter the ACLU sent Meta in the spring, the Meta patent application the complaint calls a blueprint, Meta’s own newsroom posts, the Illinois statute and its 2024 amendment, the Seventh Circuit decision that applied that amendment, and the docket.
We wanted to know what the NameTag suit actually alleges, what it can realistically recover, and where the complaint’s own numbers do not line up with the sources it cites. Face recognition is a branch of computer vision, and the case turns on whether a model trained by machine learning on photos of faces is itself a store of biometric data.
Three things stood out. Meta’s response to the lawsuit reuses 55 of the 81 words it gave WIRED in June, changing one adjective. The damages figure most outlets quote is one the Illinois legislature capped two years before this case was filed. And the complaint contains at least eight numbers that contradict either its own pages or the sources it footnotes. Our earlier report on Meta’s AI suggestions climbdown covered the Muse Image episode this complaint leans on, and our explainer on training AI models on copyrighted books covers the memorisation research it borrows.
Table of contents
- What the NameTag Lawsuit Actually Alleges
- NameTag, the Faceprint Database and Meta’s Three Statements
- Emu, Muse Image and the Billions of Photos
- The Five-Year Clock Behind the NameTag Class Period
- What NameTag Damages Look Like After the 2024 Amendment
- Eight Numbers in the Complaint That Do Not Reconcile
- How the NameTag Suit Compares With Meta’s Earlier Biometric Cases
- What Happens Next in Alvarez v. Meta
- NameTag Lawsuit FAQ
- References and Further Reading
What the NameTag Lawsuit Actually Alleges
The complaint runs to 260 numbered paragraphs, about 16,100 words by our count, and five claims. It was signed on 4 September 2026 by Justin N. Boley of Wexler Boley & Elgersma in Chicago, with three other firms on the signature block: Stephan Zouras of Chicago, Gustafson Gluek of Minneapolis and Taus, Cebulash & Landau of New York, the last two listed as pro hac vice forthcoming. The case number is 1:26-cv-10773, and the docket shows it assigned to Judge April M. Perry on 8 September.
Bloomberg Law reported the filing on the day it happened. Biometric Update published the fuller account on 7 September and hosted the PDF. WIRED’s Dhruv Mehrotra, who broke the original NameTag story in June, published his piece at 14:59 Eastern on 11 September and added Meta’s comment ten minutes later. The A.V. Club followed the same day. WIRED’s piece runs to 827 words, so most of what follows comes from the complaint itself.
The four plaintiffs and their accounts
The suit is brought by two parents and their two children, one family in each state whose law it invokes. The complaint gives an unusual amount of detail about their accounts, because under the Illinois statute the question is whether Meta obtained a person’s face without informed written consent, and the plaintiffs need to show their faces were there to be obtained.
| Plaintiff | State | Accounts | What the complaint says about the photos |
|---|---|---|---|
| Francisco Alvarez | Illinois | Two Facebook, one Instagram; about 15 and 10 years old | Posted photos of himself, friends and family every three to four months; accounts were public |
| F.M.A., a minor | Illinois | None of his own | His mother used a selfie of the two of them as her Instagram profile picture when he was about three |
| Jeremy Wahl | California | Two Facebook (personal and business), one Instagram; about 20 and 10 years old | Facebook fully public until about two years ago; Instagram always public |
| I.W., a minor | California | None of her own | Aged 10 at filing; photographed by her father from babyhood through birthdays |
The two children matter for more than sympathy. Neither has an account, so neither could have accepted any Meta terms, and the complaint uses them to make the point that Meta’s own policy says it may process images of people who do not use its products at all. That sentence, quoted from Meta’s generative AI privacy page, is one of the load-bearing exhibits of the NameTag case.
Five claims across two states
The claims split by geography. The two Illinois claims are brought under the Biometric Information Privacy Act, the statute that produced Meta’s $650 million settlement in 2021. The three California claims are a common-law right of publicity, the state’s commercial misappropriation statute and the privacy clause of the state constitution.
| Claim | Law | Who brings it | What it asks for |
|---|---|---|---|
| 1 | BIPA section 15(a), no public retention policy | Illinois class | Injunction plus $5,000 or $1,000 per violation |
| 2 | BIPA section 15(b), no notice or written release | Illinois class | Injunction plus $5,000 or $1,000 per violation |
| 3 | California common-law right of publicity | National class, California class | Damages to be set at trial, injunction |
| 4 | California Civil Code section 3344(a) | National class, California class | $750 or actual damages, plus profits, fees |
| 5 | California Constitution, right to privacy | California class | Compensatory and punitive damages, injunction |
The prayer for relief also asks for a declaration that Meta violated the two BIPA sections, an order to publish a compliant retention policy, and a permanent injunction. There is a jury demand. Jurisdiction is pleaded under the Class Action Fairness Act, which needs more than 100 class members and more than $5 million in controversy, both of which the complaint says are comfortably met.
The four uses of a face the complaint describes
Paragraph 63 is the clearest statement of the theory. It says each plaintiff’s face was used in one or more of four ways: to train the face-recognition model behind NameTag; to create a face embedding using that model; to train Meta’s generative models including Emu and Muse Image; or to produce an output from one of those models when a user submitted the image as a prompt. The first two are the NameTag half of the case. The second two are the generative half, and they are the more novel legal argument.
The complaint concedes, in paragraph 67, that Meta has never disclosed which Facebook and Instagram images it collected biometric data from and that this information is solely in Meta’s possession. WIRED made the same point in its story. What the plaintiffs offer instead is a chain of public statements and documents, which is what the next two sections examine.
NameTag, the Faceprint Database and Meta's Three Statements
The NameTag half of the complaint rests almost entirely on WIRED’s reporting, which it quotes at length, and on a New York Times story from February that it cites five times. Because the complaint quotes WIRED at length, it is worth restating what WIRED actually found and what Meta said in reply, since the reply has now been given three times in three slightly different forms.
What WIRED found on 50 million phones
On 4 June 2026 WIRED reported that code for NameTag had been quietly added to the Meta AI companion app, which had been downloaded more than 50 million times and is required for key features of the Ray-Ban and Oakley glasses. Core components were present as early as January. Three models had been deployed to phones, one to detect faces, one to crop them and one to encode them into biometric data. Recognised faces would trigger a notification. Unrecognised faces were cropped, indexed and saved to a folder marked pending.
A researcher known as Buchodi, at WIRED’s request, added a single faceprint of the philosopher Michel Foucault to the app’s gallery and triggered the pipeline. The app produced the alert “Person recognized”. Cooper Quintin of the Electronic Frontier Foundation reproduced key parts of the analysis. The next day, 5 June, Meta shipped a version of the app with the face-recognition libraries removed, which WIRED reported on 8 June. On 15 June WIRED added that Meta had licensed face-recognition software from Rank One Computing, a Denver supplier to the US Marshals Service, under a licence supporting up to 10 million facial templates.
The complaint takes one specific sentence from the June story and builds on it: that the on-device faceprint database was configured to receive updates from Meta. From that it infers a store of faceprints on Meta’s servers, and from the New York Times report that NameTag might identify people connected to the wearer or people with public Instagram accounts, it infers that the store was built from Facebook and Instagram photos. That inference is the NameTag claim in one sentence.
Meta’s statement, then and now
Meta gave WIRED an 81-word statement on 4 June and a different 81-word statement on 11 September. We compared them word for word. Fifty-five words are identical, which is 68 percent, and the closing sentence changed by exactly one adjective.
| Passage | 4 June 2026, to WIRED | 11 September 2026, to WIRED |
|---|---|---|
| Opening | “Regardless of any sensational reporting, the facts are simple: We’ve said before we’re exploring these types of features, and what you’re seeing is just evidence of that exploration.” | “This lawsuit is without merit and misrepresents our work. We’ve been transparent about how we use people’s information to build and improve our AI products. As for NameTags,” |
| Middle | “Nothing has shipped to consumers and no final decision has been made on what to do here, if anything. If we do decide to roll something out, we will take a thoughtful approach and do so with full transparency.” | Identical, 43 words unchanged |
| Closing | “One decision we can be clear about, we are not building a central face database.” | “One decision we can be clear about, we are not building a universal face database.” |
There was a third version in between. On 15 July, after chief technology officer Andrew Bosworth had described NameTag on a podcast as something that “would be a great feature”, spokesperson Ryan Daniels told WIRED that the feature “remains distinct from connecting glasses to a central database of people in the world, which is not a capability we are building.” Central became universal between July and September. Neither word is defined, and the complaint’s theory does not need a central database at all. It needs faceprints derived from Facebook and Instagram photos to exist somewhere on Meta’s side, which is a different and narrower claim.
The patent the complaint calls a blueprint
The document the plaintiffs add to WIRED’s reporting is US patent publication 2026/0142974 A1, titled “User identity verification without sharing biometric data with platforms”. Google Patents lists it as filed on 9 October 2025, published on 21 May 2026, assigned to Meta Platforms, with a single inventor, Shing Sheung Daniel Ip. Publication came 14 days before WIRED’s June story.
The application describes a network device that holds embeddings derived from a user’s stored profile images, and a communication device, which the text says may be smart glasses, that captures a face and generates embeddings locally. The privacy mechanism is that the device returns similarity scores against randomly generated embedding pairs rather than sending its biometric data up.
The complaint acknowledges this privacy-preserving design in paragraph 103 and argues that it confirms the thing that matters: Meta has built the pipeline for matching a face captured by glasses against faces it already holds from profile and tagged photos. Whether a patent application describing an architecture proves that NameTag used Facebook photos is a question a judge will have to answer. It is, at minimum, the closest the plaintiffs get to a document in Meta’s own words.
Emu, Muse Image and the Billions of Photos
The generative half of the NameTag complaint is easier to source and harder to win. It is easier to source because Meta has said in public, repeatedly, that it trains image models on public Facebook and Instagram photos. It is harder to win because the plaintiffs must persuade a court that a diffusion model’s weights contain a “scan of face geometry”, which is the statutory phrase, rather than a photograph.
The “data advantage” quote, in full
The complaint’s most quoted exhibit is a Bloomberg interview with chief product officer Chris Cox on 9 May 2024, cited to the timestamp 6:09 to 7:01. Cox said: “Our data advantage is really the public images that have been shared on Facebook and Instagram, as well as the public text. So we don’t train on private stuff, we don’t train on stuff that people shared with their friends; we do train on things that are public.”
He went on: “if you look at our image model, Emu is its name… a lot of that is because Instagram is the data set that was used to train it,” listing “imagery of people and us” among its contents.
Emu’s own paper, posted to arXiv on 27 September 2023, states that the team pre-trained a latent diffusion model on 1.1 billion image-text pairs and then fine-tuned it on only a few thousand carefully selected images. It reports an 82.9 percent win rate for the tuned model against its own pre-trained version in human preference tests. The paper does not say where the 1.1 billion pairs came from; Cox’s interview and Meta’s September 2023 newsroom post are what connect them to Instagram.
Three days of Muse Image on Instagram
The second generative exhibit is more recent. Meta launched Muse Image on 7 July 2026 as the first image model from Meta Superintelligence Labs, with an Instagram feature that let anyone generate images by @-mentioning a public account. Every public account was enrolled automatically. Three days later, on 10 July, Meta withdrew the feature with the statement that “this feature missed the mark, so it’s no longer available.” The complaint, in paragraph 42, alleges that the same capability remains reachable through the Meta AI website and app, an allegation Meta has not addressed in its statement on the suit.
The complaint uses Muse Image in two ways. First, as evidence that users did not consent to and do not want their images used. Second, as the event that starts the discovery clock for the generative claims, which we return to below. Meta’s 10 July statement ran to 67 words by our count when we covered it in September, and the complaint quotes every one of them.
The memorisation argument
To turn a trained model into a biometric store, the complaint borrows from two research papers. Carlini and colleagues showed in January 2023 that diffusion models memorise and regenerate individual training images, including photographs of identifiable people, and memorise about twice as much as comparable GANs. Cooper and colleagues showed in a paper updated in July 2026 that Llama 3.1 70B has memorised some books almost entirely, with a near-verbatim copy of Harry Potter and the Sorcerer’s Stone extractable from its first few words.
Against that, the complaint quotes Meta’s own brief in the Kadrey copyright case, filed 24 March 2025, which told the court that “nothing of the training data itself remains” in Llama. The plaintiffs’ point is that Meta cannot say both things.
If a model retains enough of a training image to reproduce a face, the identity-relevant measurements it encodes are, on the complaint’s reading, biometric identifiers under the Illinois statute whether they sit in an embedding or in model parameters. The statute defines a biometric identifier to include a “scan of face geometry”, and no court has yet said whether a latent representation inside a generative AI model qualifies. That is the question on which the generative half of NameTag will rise or fall, and it will be litigated on a motion to dismiss before any class is certified.
The Five-Year Clock Behind the NameTag Class Period
The class period in the NameTag complaint runs from 4 September 2021 to the present. That start date is not arbitrary. It is exactly five years before the filing date, and five years is the limitation period the Illinois Supreme Court fixed for every BIPA claim in Tims v. Black Horse Carriers on 2 February 2023. The plaintiffs have drawn the class as wide as the statute allows and then argued, separately, that the clock should not have started at all until mid-2026.
Three classes, one date
| Class | Definition | Claims | Size pleaded |
|---|---|---|---|
| Illinois | All persons in Illinois whose image was uploaded to Facebook, Instagram or a Meta generative model at inference | 1 and 2 (BIPA) | “Hundreds of thousands” |
| National | All persons in the United States, same three routes | 3 and 4 (California law) | “In the millions” |
| California | All persons in California, same three routes | 5, and 3 and 4 in the alternative | “Hundreds of thousands” |
Two features of the definitions are worth noticing. The classes are defined by whose image was uploaded, not by who uploaded it, so a person who never had an account is in scope if a friend posted their photo. And the third route, an image submitted to a Meta model as a prompt, sweeps in people whose faces were never posted on either platform but were fed to Meta AI by someone else.
The complaint supports its size estimates with the figures of over 250 million Facebook users and over 160 million Instagram users in the United States, and with Emu’s 1.1 billion training pairs, which works out to about 2.7 pairs per US user across both platforms if one were to spread them evenly, which of course nobody claims they were.
Why 4 September 2021
Tims settled a long argument about whether some BIPA claims carried a one-year privacy limitation period and others the five-year catch-all. The Illinois Supreme Court chose five years for all of section 15. A complaint filed on 4 September 2026 can therefore reach conduct back to 4 September 2021 without any help from tolling doctrines, and that is what the NameTag class period does.
The date also sits after Meta’s November 2021 announcement that it was shutting down Facebook’s face-recognition system and deleting more than a billion people’s templates. The class period therefore begins two months before the deletion Meta publicised, which means the NameTag plaintiffs are not suing over the old tag-suggestion system. They are suing over what they say was built afterwards.
The tolling argument
Section D of the complaint pleads delayed discovery, equitable tolling and fraudulent concealment in the alternative. Its argument is that nobody could have known Meta was deriving faceprints from platform photos until WIRED’s June 2026 report, and nobody could have known the scope of generative training until the Muse Image backlash in July 2026, so the limitation period did not start until mid-2026. That matters less for the Illinois claims, which already have five years, than for the California claims, where the limitation periods are shorter and the plaintiffs need the clock to have started recently.
The takeaway is that the intervals in this story are short. WIRED’s report to Meta’s code removal took one day; the report to the complaint took 92.
What NameTag Damages Look Like After the 2024 Amendment
Every report on the NameTag suit quotes the same two numbers: $5,000 for each intentional or reckless violation of BIPA and $1,000 for each negligent one. Those figures are in the statute and in the prayer for relief. What the reports leave out is that Illinois changed what counts as a violation on 2 August 2024, and that the Seventh Circuit held on 1 April 2026 that the change applies to cases already pending. This case was filed 763 days after the amendment took effect, so there is no argument about retroactivity to be had.
One recovery per person, not per photo
Senate Bill 2979, signed by Governor Pritzker on 2 August 2024 and effective the same day, added a rule that a private entity which collects the same biometric identifier from the same person by the same method more than once has committed a single violation, entitling that person to at most one recovery. Before the amendment, the Illinois Supreme Court had held in Cothron v. White Castle that a claim accrued on every scan, which is what turned a fingerprint time clock into a nine-figure exposure. The amendment ended per-scan accrual.
In Clay v. Union Pacific Railroad, decided on 1 April 2026, the Seventh Circuit held that the amendment is remedial, touching only the damages available rather than the liability rules, and therefore reaches cases filed before August 2024. For a case filed in September 2026, that ruling is belt and braces. For the NameTag class it means the arithmetic is per person per method, not per photo: a member who appears in a thousand Instagram photos and one faceprint is, for damages purposes, one violation per collection method.
| Remedy | Source | Unit | Applies to |
|---|---|---|---|
| $5,000 liquidated or actual damages | 740 ILCS 14/20(2) | Per intentional or reckless violation, one recovery per person per method since 2024 | Illinois class |
| $1,000 liquidated or actual damages | 740 ILCS 14/20(1) | Per negligent violation, same single-recovery rule | Illinois class |
| Injunction and a public retention policy | 740 ILCS 14/20(4) and 15(a) | Destroy data within 3 years of last interaction | Illinois class |
| $750 or actual damages, plus attributable profits | Cal. Civ. Code 3344(a) | Per person; minors need parental consent | National and California classes |
| Compensatory and punitive damages | California Constitution, art. I, s. 1 | Set at trial | California class |
What the earlier Illinois case paid
The reference point is In re Facebook Biometric Information Privacy Litigation, the tag-suggestions case. Judge James Donato of the Northern District of California gave final approval to a $650 million settlement on 26 February 2021, for a class of up to six million Illinois users, and said each claimant would receive at least $345. Meta had originally offered $550 million in January 2020 and raised it to $650 million that July. Texas then obtained $1.4 billion in July 2024 under its own biometric statute, paid over five years, without a class.
The complaint’s paragraph 13 lists Meta’s privacy penalties over seven years and says they total more than $10 billion. The five items it names add up to $9.075 billion, which is not a contradiction since the list is introduced with “including”, but the reader is left to supply the remaining $925 million.
The arithmetic the coverage implies and the arithmetic the statute allows
A reader who multiplies $5,000 by “millions” gets a number in the tens of billions, and that is the impression the headline figures leave. The statute no longer supports it. Under the single-recovery rule the ceiling is the per-person figure multiplied by the Illinois class, since the statutory damages apply only to the Illinois claims.
On the complaint’s own “hundreds of thousands” description, every 100,000 Illinois members represent $100 million at the negligent rate or $500 million at the intentional rate, before any court weighs whether liquidated damages of that size are proportionate, which BIPA courts have said they may do. The 2021 settlement, at $650 million for up to six million people, came in at roughly $108 per class member gross. Anyone reading the NameTag numbers should hold both figures in mind.
Eight Numbers in the Complaint That Do Not Reconcile
We checked every figure in the complaint against the source it footnotes or against the complaint’s other pages. The core allegations survive. The framing does not, entirely. These are the eight discrepancies we found, in page order.
| Where | The complaint says | What the source or the complaint itself says |
|---|---|---|
| Para 9 | “In fiscal year 2024, Meta generated approximately $200.97 billion in revenue and $60.46 billion in net income” | Those are the 2025 figures in Meta’s 28 January 2026 release ($200,966m and $60,458m). Fiscal 2024 was $164,501m and $62,360m |
| Para 9 and 14 | Market value “exceeded $1.6 trillion”, then “the $1.3 trillion company” five paragraphs later | Two different valuations for the same company on consecutive pages |
| Para 13 | “a $650 settlement stemming from violations of the Illinois Biometric Information Privacy Act” | $650 million; the word is missing |
| Para 82 | An April 2026 open letter “co-signed by 75 local, state, and national organizations” | The letter’s first sentence says “the undersigned 77”, and 77 names follow the signature; WIRED wrote “more than 70” |
| Para 123 and 133 | Emu pre-trained on “1.1 billion image-text pairs”, then “at least 1.2 billion image-text pairs” ten paragraphs later | The Emu paper says 1.1 billion; paragraphs 68 and 197 also say 1.1 billion |
| Para 126 | “at least three distinct Emu-based models: (1) Emu… (2) Emu Edit… and (3) Emu Edit” | The InfoQ article it cites covers Emu Video and Emu Edit; the third item duplicates the second |
| Para 46 | Violations of BIPA “section 15(a) to (c)” | Claims 1 and 2 plead only 15(a) and 15(b); no 15(c) claim appears |
| Claims 1 and 2 headings | Brought by “Francisco Alvarez and I.W., a minor proceeding by and through his guardian Francisco Alvarez” | I.W. is Jeremy Wahl’s daughter in California; the Illinois minor is F.M.A. The prayer for relief names F.M.A. correctly |
Two smaller slips are worth a line. WIRED’s reporter is spelled “Mehrorta” in both footnotes that cite him. And the $10 billion total we covered above is short by $925 million on the items listed. None of these touches the numbers the case actually depends on: the 50 million downloads, the 1.1 billion training pairs, the January date for the NameTag code, and the $5,000 and $1,000 statutory figures all match their sources.
What the slips have in common
Every one of the eight is in the narrative sections, the introduction and the factual background, rather than in the class definitions, the claims or the prayer for relief. That is where amended complaints usually tidy up, and an amended complaint is routine at this stage of a class action. It is also a reminder that a complaint is an advocacy document. The NameTag allegations are serious and specifically sourced, and they are also drafted to be read by a jury.
How the NameTag Suit Compares With Meta's Earlier Biometric Cases
This is at least the fourth time Meta’s handling of faces has ended up in a courtroom or an attorney general’s office, and the second lawsuit this year about its glasses. The table below sets the NameTag case beside the three it is most often compared with.
| Case | Court and filing | Technology at issue | Outcome or status |
|---|---|---|---|
| In re Facebook Biometric Information Privacy Litigation | N.D. California, filed 2015 | Facebook tag suggestions and face templates | $650m settlement approved 26 Feb 2021; opt-in required; templates deleted |
| Texas v. Meta Platforms | Texas state court, filed 2022 by the attorney general | Face geometry captured from uploaded photos, Texas CUBI statute | $1.4bn over five years, announced 30 Jul 2024; first case ever under CUBI |
| Bartone v. Meta Platforms and Luxottica | N.D. California, filed 4 Mar 2026, amended Sep 2026 | Glasses recordings sent to human reviewers in Kenya; bystander class added | Pending; Meta says it “will fight” the claims |
| Alvarez v. Meta Platforms (NameTag) | N.D. Illinois, filed 4 Sep 2026 | Faceprints for NameTag and training of Emu and Muse Image from platform photos | Pending before Judge Perry; Meta says it is “without merit” |
Same statute, different technology
The 2015 case and the NameTag case share a statute and a defendant and almost nothing else. The earlier system scanned photos to suggest tags, and Facebook admitted the templates existed; the argument was about consent. Here Meta denies that any consumer product exists and the plaintiffs must first prove the faceprints are there. The NameTag case also adds the generative claims, which have no precedent in the 2015 litigation because Emu did not exist until 2023.
The other glasses case
Bartone is the case most likely to be confused with NameTag. It was filed on 4 March 2026 by the Clarkson Law Firm in San Francisco on behalf of glasses purchasers, alleging that footage marketed as private was sent to contractors in Kenya to label for AI training. In September the plaintiffs amended to add a class of bystanders recorded without consent. It concerns what the glasses record. The NameTag case concerns what Meta already held before anyone put the glasses on.
What the 2021 shutdown promised
On 2 November 2021 Jerome Pesenti, then Meta’s vice president of artificial intelligence, announced that Facebook would shut down its face-recognition system, delete the templates of more than a billion people, and limit future use to narrower cases such as identity verification. He noted that more than a third of daily active users had opted in. The NameTag complaint’s class period starts two months before that post. If the plaintiffs are right, the narrower cases turned out to include a consumer wearable; if Meta is right, nothing was ever built beyond an exploration. The 2021 promise is the yardstick either way.
What Happens Next in Alvarez v. Meta
The docket is short. The complaint and civil cover sheet were filed on 4 September, the case was assigned to Judge April M. Perry on 8 September, and an attorney appearance was entered the same day. Meta had not been served as of the entries we could see, and its statement to WIRED is the only response on record.
The docket so far
| Date | Entry | Note |
|---|---|---|
| 4 Sep 2026 | Class action complaint, 66 pages, jury demand | Free on CourtListener as document 1 |
| 8 Sep 2026 | Case assigned to Judge April M. Perry | Clerk’s notice the same day |
| 8 Sep 2026 | Attorney appearance, document 4 | Plaintiffs’ counsel |
| 11 Sep 2026 | WIRED story; Meta statement added 15:09 Eastern | Not a docket event |
The question a judge must answer first
Meta’s first move will almost certainly be a motion to dismiss, and the NameTag complaint anticipates it. The plaintiffs will need to show they have plausibly alleged that Meta possesses biometric identifiers derived from their faces, not merely that it could. For the NameTag half that turns on WIRED’s finding that the on-device database was configured to receive updates from Meta, plus the patent. For the generative half it turns on whether a court accepts that model parameters trained on faces are a scan of face geometry. That is a genuinely open question, and whichever way Judge Perry answers it will be cited in every AI training case that follows.
What Meta has said it will do
Meta’s position, in its statement to WIRED, is that “nothing has shipped to consumers and no final decision has been made”. Bosworth’s July description of a NameTag that recognises only people the wearer has asked the glasses to remember is the narrowest version of the feature Meta has offered in public.
If that is the product that eventually ships, the faceprint database would be built by the wearer, not from Instagram, and the NameTag half of this case would be about what was built and deleted in 2026 rather than about anything on sale. The generative half does not depend on the glasses at all, which is probably why it is there.
NameTag Lawsuit FAQ
Who is suing Meta over NameTag?
Francisco Alvarez and his minor son from Illinois, and Jeremy Wahl and his 10-year-old daughter from California, on behalf of proposed Illinois, national and California classes. They are represented by Wexler Boley & Elgersma, Stephan Zouras, Gustafson Gluek and Taus, Cebulash & Landau. The case is Alvarez v. Meta Platforms, Inc., No. 1:26-cv-10773, in the Northern District of Illinois.
Was NameTag ever switched on for users?
No. WIRED found functional but inactive face-recognition code in the Meta AI app in June 2026, present since January. Meta removed it on 5 June, the day after the report, and says no final decision has been made on any such feature.
What does the NameTag complaint say Meta did with Facebook and Instagram photos?
It alleges Meta used them to train the face-recognition model behind NameTag, to generate faceprints for NameTag to match against, and to train the Emu and Muse Image generative models, all without the notice and written consent the Illinois statute requires.
How much could the NameTag case be worth?
BIPA provides $1,000 per negligent and $5,000 per intentional or reckless violation, but since the 2 August 2024 amendment a person can recover only once per collection method, and the Seventh Circuit confirmed in April 2026 that the rule applies retroactively. The realistic ceiling is a per-person figure multiplied by the Illinois class, plus whatever the California claims add.
How does this differ from the Bartone glasses lawsuit?
Bartone, filed in California in March 2026, is about what the glasses record and where the footage goes. The NameTag case is about what Meta allegedly extracted from photos already on its platforms, before any glasses were involved.
What has Meta said?
That the lawsuit “is without merit and misrepresents our work”, that nothing has shipped, and that it is “not building a universal face database”. In June the last phrase read “central face database”.
References and Further Reading
Meta Sued Over Training Data for Its AI and Face-Recognition Systems (WIRED, 11 September 2026)
Alvarez v. Meta Platforms, Inc., Class Action Complaint, N.D. Ill. No. 1:26-cv-10773 (CourtListener)
Alvarez v. Meta Platforms, Inc. docket (CourtListener)
Meta sued over alleged facial recognition training for smart glasses (Biometric Update)
Meta Sued Over Smart Glasses’ Collection of User Biometric Data (Bloomberg Law)
Meta Tapped a Pentagon Supplier to Prototype Face Recognition for Its Glasses (WIRED, 15 June 2026)
Here’s the Truth About Whether Meta’s NameTag Face Recognition Tech ‘Exists’ (WIRED, 15 July 2026)
Emu: Enhancing Image Generation Models Using Photogenic Needles in a Haystack (arXiv)
Extracting Training Data from Diffusion Models (Carlini et al., arXiv)
Meta removes controversial AI feature on Instagram after backlash (TechCrunch)
An Update On Our Use of Face Recognition (Meta Newsroom, 2 November 2021)
Attorney General Ken Paxton Secures $1.4 Billion Settlement with Meta (Texas Attorney General)
Governor Pritzker Signs Amendment Limiting BIPA Damages to a Single Recovery (Seyfarth Shaw)
Meta Reports Fourth Quarter and Full Year 2025 Results (Meta Investor Relations)
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.