Anatomy of the $1.4 Billion Settlement for CUBI Violations
1. The Financial Architecture of the Settlement
The agreement mandates that Meta pay the State of Texas a total of $1. 4 billion. This figure represents the largest privacy settlement ever obtained by a single state attorney general, surpassing the $390 million multi-state settlement with Google in 2022. Unlike class-action lawsuits where damages are distributed among millions of claimants, these funds are directed to the state treasury. The magnitude of the fine reflects the statutory penalties available under Texas law, which allow for civil penalties of up to $25, 000 per violation. With millions of Texans using Facebook, the chance liability exceeded hundreds of billions of dollars, providing the state with substantial use during negotiations.
2. The Five-Year Payment Schedule
The settlement terms permit Meta to disburse the $1. 4 billion over a five-year period. This structured payout mitigates the immediate liquidity impact on the company while ensuring a steady revenue stream for the state. The installment was due 30 days after the judicial approval of the settlement.
| Payment Tranche | Due Date | Amount |
|---|---|---|
| Installment | August 2024 (30 days post-agreement) | $500, 000, 000 |
| Second Installment | 2025 | $225, 000, 000 |
| Third Installment | 2026 | $225, 000, 000 |
| Fourth Installment | 2027 | $225, 000, 000 |
| Fifth Installment | 2028 | $225, 000, 000 |
| Total | 2024, 2028 | $1, 400, 000, 000 |
3. The Statutory Weapon: Texas CUBI Act of 2009
The lawsuit relied exclusively on the Texas Capture or Use of Biometric Identifier Act (CUBI), enacted in 2009. While Illinois’s Biometric Information Privacy Act (BIPA) frequently garners attention for its private right of action, allowing citizens to sue directly, CUBI restricts enforcement power solely to the Texas Attorney General. The statute (Tex. Bus. & Com. Code § 503. 001) prohibits companies from capturing a biometric identifier, defined as a retina or iris scan, fingerprint, voiceprint, or record of hand or face geometry, for a “commercial purpose” unless the company:
- Informs the individual before capturing the biometric identifier; and
- Receives the individual’s consent to capture the biometric identifier.
Texas Attorney General Ken Paxton argued that Meta failed on both counts. The state contended that the “Tag Suggestions” feature automatically captured facial geometries from uploaded photos without explicit, informed consent from the subjects in those photos.
4. The “Tag Suggestions” method and DeepFace
The core of the dispute involved Meta’s “Tag Suggestions” feature, powered by an artificial intelligence system known as DeepFace. Introduced in 2011, this technology analyzed photos uploaded to the platform to identify faces. The process functioned in three steps:
Detection: The system scanned an uploaded image to detect the presence of a face.
Alignment and Measurement: The software mapped the geometry of the face, calculating the distances between key features (eyes, nose, ears) to create a unique numerical string or “template.”
Matching: This template was compared against a database of existing user templates to suggest a “tag.”
The state alleged that this process occurred billions of times on photos uploaded by Texans. Crucially, the system analyzed faces of users who had not explicitly opted in, as well as non-users who appeared in photos uploaded by others. The state argued this constituted the capture of biometric identifiers without consent.
5. The “Commercial Purpose” Argument
A serious legal battleground was the definition of “commercial purpose.” CUBI only penalizes the capture of biometrics for commercial gain. Meta historically argued that improving user experience (tagging friends) was not a commercial activity in the direct sense of selling data. Texas prosecutors countered that Meta used the biometric data to train and improve its artificial intelligence capabilities (DeepFace). By refining its AI on the faces of Texans, Meta enhanced its core product, which in turn drove user engagement and advertising revenue. The state successfully positioned the training of the AI algorithm as a direct commercial benefit derived from the unauthorized data.
6. Comparative Analysis: Texas CUBI vs. Illinois BIPA
The $1. 4 billion Texas settlement offers a sharp contrast to the $650 million class-action settlement Meta agreed to in Illinois in 2021 over similar violations.
Plaintiff Structure: The Illinois case was a class action suit where the settlement funds were divided among approximately 1. 6 million claimants, resulting in individual checks of roughly $397. In Texas, the Attorney General sued on behalf of the state, meaning the entire $1. 4 billion goes to the state government, not individual Facebook users.
Penalty Calculation: The Illinois settlement was capped by the size of the settlement fund relative to the class. In Texas, the state threatened to seek the maximum statutory penalty of $25, 000 per violation for millions of users, creating a liability exposure that theoretically reached into the trillions. This use forced a higher settlement figure even without a trial.
7. The “Knowing” Violation and Intent
The petition filed by Texas in the 71st Judicial District Court of Harrison County emphasized that Meta’s violations were “knowing.” The state pointed to the ongoing litigation in Illinois (which began in 2015) as evidence that Meta was fully aware of the legal risks associated with biometric data collection yet continued the practice in Texas until late 2021. By establishing that Meta knew of the biometric privacy requirements and continued its practices, Texas positioned itself to seek maximum penalties. The state argued that Meta did not disable the feature in Texas even after facing scrutiny elsewhere, thereby exposing the company to higher damages under the Deceptive Trade Practices Act (DTPA), which was filed alongside the CUBI claims.
8. Data Destruction and Operational Changes
While the financial penalty is the headline, the settlement reinforces the operational changes Meta made prior to the agreement. In November 2021, Meta announced it would shut down the Face Recognition system on Facebook and delete the facial recognition templates of more than one billion people globally. The Texas settlement formalizes this obligation within the state. It ensures that any biometric data previously collected from Texans through the Tag Suggestions feature is permanently destroyed and cannot be resurrected or used to train future models without strict compliance with CUBI. This bars Meta from re-enabling facial recognition features in Texas without a clear, opt-in consent method that meets the 2009 law’s standards.
9. The Role of Outside Counsel
The Texas Attorney General’s office did not litigate this case alone. The state retained two prominent private law firms, Keller Postman and McKool Smith, to lead the aggressive litigation strategy. This arrangement is notable because it allows the state to match the legal resources of a trillion-dollar corporation. The use of private counsel in state enforcement actions is a growing trend, allowing states to pursue complex technical litigation that requires specialized expertise in data privacy and algorithmic accountability.
10. Jurisdiction and Venue Strategy
The lawsuit was filed in a state district court in Harrison County, Texas, rather than a federal court or a major urban center like Austin or Dallas. This venue selection was strategic. State courts in Texas are frequently viewed as favorable venues for plaintiffs in high- liability cases. By keeping the case in state court and anchoring it on a specific Texas statute (CUBI) that has no federal equivalent, the state prevented Meta from removing the case to federal court, where standing requirements (Article III standing) are frequently stricter and more favorable to tech defendants.
11. for Future AI Training
The settlement establishes a de facto price on the use of biometric data for AI training. By accepting a $1. 4 billion penalty, the industry has a benchmark for the cost of non-compliance with biometric statutes. It signals that “commercial purpose” includes the internal training of AI models, not just the direct sale of data. This precedent affects not only social media companies any enterprise capturing voice, face, or gait data to refine machine learning algorithms.
12. The “Bystander” Problem
A specific focus of the Texas investigation was the capture of data from “bystanders” or non-users. When a Facebook user uploaded a group photo, the DeepFace algorithm scanned every face in the image to find matches. This meant Meta was measuring the facial geometry of people who never signed up for Facebook and never consented to its Terms of Service. While the settlement resolves the state’s claims, the legal theory that non-users have viable claims under biometric laws remains a potent risk for companies deploying computer vision in public or semi-public spaces.
The DeepFace Algorithm and Unconsented Face Geometry Harvesting
The Architecture of DeepFace
In June 2014, Facebook’s Artificial Intelligence Research (FAIR) unit published a paper titled “DeepFace: Closing the Gap to Human-Level Performance in Face Verification.” This document detailed the deployment of a nine- deep neural network comprising more than 120 million parameters. Unlike previous iterations of facial recognition that relied on simple 2D feature matching, DeepFace employed a process called “frontalization.” The algorithm constructs a 3D model of a face from a 2D image, rotates it to a forward-facing alignment, and then extracts a numerical signature.
The system outputs a 4, 096-dimensional feature vector, a string of numbers representing the unique topography of a human face. In standardized tests using the “Labeled Faces in the Wild” (LFW) dataset, DeepFace achieved an accuracy rate of 97. 35%. For comparison, human beings performing the same task achieve an accuracy of 97. 53%. This technological leap rendered the platform’s automated surveillance capabilities indistinguishable from human perception.
| Evaluator | Methodology | Accuracy Rate | Error Rate |
|---|---|---|---|
| Human Being | Visual Cortex / Pattern Recognition | 97. 53% | 2. 47% |
| DeepFace (Meta) | 9- Deep Neural Network | 97. 35% | 2. 65% |
| Previous State-of-the-Art | Conventional Computer Vision | ~70-75% | ~25-30% |
The “Tag Suggestions” method
Meta integrated this algorithm into the user experience through a feature labeled “Tag Suggestions.” When a user uploaded a photograph, the system automatically scanned the image for faces. Before the user took any action to identify the subjects, DeepFace had already detected the facial regions, aligned them, and computed their biometric signatures. It then compared these new signatures against a massive centralized database of stored user templates.
If the algorithm found a match exceeding a specific confidence threshold, it prompted the uploader to tag the individual. While the user interface presented this as a convenience tool to save time, the backend process constituted a non-consensual extraction of biometric data. The system did not distinguish between users who had consented to biometric capture and those who had not; it scanned the geometry of every detected face to determine if a match existed.
Violation of Texas CUBI Statutes
The core of the State of Texas’s legal argument rested on the definition of a “biometric identifier” under the Capture or Use of Biometric Identifier Act (CUBI). The statute explicitly defines this term to include a “record of face geometry.”
“A person may not capture a biometric identifier of an individual for a commercial purpose unless the person: (1) informs the individual before capturing the biometric identifier; and (2) receives the individual’s consent to capture the biometric identifier.”
, Texas Business and Commerce Code § 503. 001
Court filings reveal that Meta failed to obtain this informed consent from approximately 20. 5 million Texans. The violation occurred at the moment of the scan, the creation of the 4, 096-dimensional vector, rather than the application of the visible tag. Because the “Tag Suggestions” feature was enabled by default, users were automatically subjected to this biometric serialization without an opt-in method. also, the system captured the facial geometries of non-users (individuals without Facebook accounts) who appeared in photos uploaded by others, creating “orphan” biometric records for individuals who had no relationship with the platform.
The 2021 Shutdown and Data Deletion
Facing mounting regulatory pressure and class-action litigation in Illinois (BIPA), Meta announced on November 2, 2021, that it would shut down the Face Recognition system. Jerome Pesenti, then-VP of Artificial Intelligence, confirmed the company would delete the facial recognition templates of more than one billion people. This mass deletion event served as a tacit admission of the liability generated by holding such a vast repository of sensitive biometric data. yet, the Texas lawsuit proceeded, arguing that the deletion of data in 2021 did not absolve the company of liability for the decade of unauthorized capture and commercial use that preceded it.
Operational Timeline of the Tag Suggestions Feature 2011 to 2021
The Rise and Fall of “Tag Suggestions”
The operational core of the Texas lawsuit centers on a specific feature Meta (then Facebook) introduced in 2011: “Tag Suggestions.” While marketed as a user convenience tool designed to streamline the photo-tagging process, court documents filed by the Texas Attorney General in 2022 reveal the feature functioned as a sophisticated biometric data harvesting system. For over a decade, this system automatically scanned uploaded photographs, identified faces, and suggested names by comparing facial geometries against a massive proprietary database.
The serious legal friction point in Texas arose from the enrollment method. From its inception until 2019, the feature operated on an opt-out basis. When a Texas resident created a Facebook account, the platform automatically enabled facial recognition by default. Consequently, the system captured the biometric identifiers of millions of Texans without the “informed consent” required by the state’s Capture or Use of Biometric Identifier Act (CUBI). The Texas petition estimates that this default setting affected approximately 20. 5 million Texans, leading to the capture of biometric data billions of times as users uploaded photos to the platform.
The DeepFace Algorithm
The efficacy of Tag Suggestions relied on “DeepFace,” a deep learning facial recognition system Facebook researchers published details on in 2014. By 2015, reports confirmed that DeepFace operated with an accuracy rate of 97. 35%, closely method the human-level accuracy of 97. 53%. Unlike earlier facial detection software that simply identified the presence of a face, DeepFace generated a unique digital signature, or “template”, for each user.
This template relied on measuring the distances between specific facial features, such as the eyes, nose, and ears, to create a mathematical representation of the user’s face. The 2022 Texas lawsuit alleged that Meta used these templates not only to suggest tags also to train and improve its artificial intelligence capabilities, constituting a “commercial purpose” under CUBI statutes. The system processed photos even for users who had not uploaded them, scanning group shots and background faces to check for matches against its enrolled database.
Regulatory Pivots and the 2021 Shutdown
Following intense scrutiny and a record $5 billion FTC settlement regarding privacy practices in 2019, Meta altered the operational structure of its facial recognition tools. In September 2019, the company announced it would deprecate the “Tag Suggestions” setting, replacing it with a broader “Face Recognition” toggle. Crucially, this update shifted the enrollment model from opt-out to opt-in for new users, though it required existing users to actively disable the feature if they had been previously enrolled by default.
The operational timeline concluded abruptly on November 2, 2021. Jerome Pesenti, then-Vice President of Artificial Intelligence at Meta, announced the company would shut down the Face Recognition system entirely. Pesenti “growing concerns about the use of this technology as a whole” and the absence of clear regulatory rules as the primary drivers for the decision. As part of this decommissioning process, Meta committed to deleting the facial recognition templates of more than one billion people globally. This cessation of data collection capped the period of alleged violations in the Texas lawsuit, although the state argued that the prior decade of non-consensual data capture warranted the statutory penalties that eventually led to the $1. 4 billion settlement in 2024.
Chronology of the Tag Suggestions Feature
| Year | Operational Event | Impact on Texas Litigation |
|---|---|---|
| 2011 | Global rollout of “Tag Suggestions.” | Feature enabled by default; alleged start of CUBI violations. |
| 2014-2015 | DeepFace integration achieves 97. 35% accuracy. | Established the high-fidelity biometric nature of the data captured. |
| 2019 | Tag Suggestions replaced by “Face Recognition” setting. | Shift to opt-in model for new users; legacy data retention continued. |
| 2021 | Meta announces shutdown of facial recognition system. | Deletion of 1 billion templates; end of the alleged violation period. |
| 2022 | Texas AG files suit citing 2011-2021 operations. | Formal legal challenge to the decade-long data collection practices. |
Comparative Metrics Between Texas CUBI and Illinois BIPA Payouts
Statutory use: The $25, 000 Multiplier
The primary driver of the $1. 4 billion figure is the penalty structure codified in the Texas Business and Commerce Code. Under CUBI, the state does not need to prove actual damages to individual citizens. The violation itself, capturing a biometric identifier without consent, triggers the penalty. In the Illinois class action, the plaintiffs had to negotiate a settlement based on the risks of trial and the difficulty of proving “reckless” conduct to unlock the higher $5, 000 tier. In contrast, the Texas Attorney General entered negotiations with a statutory ceiling that was theoretically in the trillions of dollars. With approximately 20. 5 million Texans chance affected by the “Tag Suggestions” feature, and multiple violations per user over a decade, the maximum exposure for Meta was existential. This use forced a settlement value that reflects a significant percentage of the company’s quarterly revenue rather than a simple cost of doing business.
| Feature | Texas CUBI | Illinois BIPA |
|---|---|---|
| Enforcement method | Attorney General Only | Private Right of Action (Class Action) |
| Penalty Per Violation | Up to $25, 000 | $1, 000 (Negligent) / $5, 000 (Reckless) |
| Statute of Limitations | No express limit (interpreted broadly) | 5 Years (confirmed in Tims v. Black Horse) |
| Settlement Recipient | State General Revenue Fund | Individual Class Members |
| Meta Settlement Amount | $1. 4 Billion | $650 Million |
The Illinois Benchmark: In re Facebook
The Illinois settlement serves as the control group for this analysis. Finalized in February 2021, the In re Facebook case involved 1. 6 million valid claimants. The total settlement fund was $650 million. After legal fees and administrative costs, each claimant received approximately $397. This remains one of the highest per-person payouts in a consumer privacy class action. The Texas settlement operates on a different of efficiency. There was no claims process. There was no need to locate class members or verify their residency. The state acted as the sole plaintiff. Consequently, the entire $1. 4 billion moves directly to the state, bypassing the fragmentation of funds that occurs in class actions. While Illinois residents received direct compensation for the violation of their privacy, Texas residents received no direct payments. The “benefit” in Texas is indirect, flowing into the state’s general budget, whereas the “benefit” in Illinois was a tangible check in the mail.
Comparative Settlement Valuation
When analyzing the broader of biometric privacy settlements, the Texas CUBI victory appears even more distinct. Other major technology companies have settled BIPA claims in Illinois, none have method the billion-dollar threshold. The structural limitations of BIPA class actions, specifically the need to certify a class and the pressure to settle before trial, frequently result in lower aggregate numbers. The following list details major BIPA settlements in Illinois between 2020 and 2025. These figures highlight how the Texas CUBI statute functions as a “super-weapon” in comparison. * Google (Illinois BIPA): Settled for $100 million in 2022 regarding the “Face Grouping” feature in Google Photos. * TikTok (Illinois BIPA): Settled for $92 million in 2022 over allegations of collecting facial geometry for filters and ad targeting. * Snapchat (Illinois BIPA): Settled for $35 million in 2022 regarding its “Lenses” and filters. * Instagram (Illinois BIPA): Settled for $68. 5 million in 2023. Note that this is separate from the main Facebook settlement. In May 2025, Texas further validated the power of CUBI by securing a $1. 375 billion settlement with Google over similar biometric allegations. This second billion-dollar recovery confirms that the Meta settlement was not a fluke a reproducible result of the state’s aggressive legal strategy and the draconian penalty caps within the CUBI statute.
The Role of Outside Counsel
A serious factor in the between the Texas and Illinois outcomes is the legal representation model. In the Illinois In re Facebook case, class counsel (Edelson PC and others) litigated for years to secure the $650 million fund. Their fees were determined by the court as a percentage of the fund, around 15% to 30%. In the Texas case, Attorney General Ken Paxton retained outside private counsel, specifically the firms Keller Postman and McKool Smith, to litigate on behalf of the state. This arrangement allowed the state to deploy the resources of high-end litigation firms without upfront cost. The fee structure for these firms was substantial. Public records indicate the firms billed approximately $136 million in fees, a figure derived from a contingency arrangement capped by state law. This aggressive litigation posture, backed by private sector resources, allowed Texas to match Meta’s legal defense team motion for motion.
Data Visualization: The Settlement Hierarchy
The visual gap between the Texas and Illinois settlements illustrates the financial impact of the CUBI statute. The chart represents the total settlement amounts paid by major tech companies for biometric violations in each jurisdiction.
Chart Note: The following data points represent the total gross settlement funds before attorneys’ fees or administrative costs.
Texas (Meta): $1, 400, 000, 000
Texas (Google): $1, 375, 000, 000
Illinois (Meta): $650, 000, 000
Illinois (Google): $100, 000, 000
Illinois (TikTok): $92, 000, 000
Outcomes for the Consumer
The most significant difference between the two jurisdictions lies in the end result for the citizen. In Illinois, the BIPA statute is designed as a consumer protection tool that incentivizes private enforcement. The $397 checks sent to 1. 6 million Illinois residents represented a tangible transfer of wealth from the corporate entity to the aggrieved individuals. It served as both a punishment to the company and a remediation to the user. Texas CUBI functions differently. It is a regulatory enforcement tool. The $1. 4 billion payment is a civil penalty paid to the sovereign. The individual Texans whose faces were scanned without consent did not receive a portion of the settlement. The funds were deposited into the General Revenue Fund, where they are subject to appropriation by the Texas Legislature. This distinction raises questions about the purpose of privacy laws: are they meant to compensate the victim, or are they meant to punish the violator? The Texas model prioritizes the latter. By treating the violation as an offense against the state rather than a tort against the individual, Texas maximized the financial penalty. The state argued that the sheer volume of violations, billions of photos uploaded by millions of Texans over a decade, warranted a maximum-penalty method. Meta, facing the prospect of a trial where the statutory math could lead to bankruptcy-level judgments, chose to pay the $1. 4 billion premium to close the matter.
The “Opt-In” vs. “Parens Patriae” Factor
In the Illinois class action, the participation rate was a key metric. Only about 22% of eligible Illinois Facebook users filed a claim. This is considered a high participation rate for a class action, yet it still means that nearly 80% of affected users received nothing. The total settlement amount was negotiated based on the size of the class and the strength of the claims, the final payout was limited by the number of people who actually took the step to file. Texas utilized the doctrine of parens patriae, which allows the Attorney General to sue on behalf of all citizens. There was no “opt-in” requirement. The state represented every single Texan with a Facebook account, regardless of whether that person knew about the lawsuit or cared to participate. This total-coverage method allowed the state to calculate damages based on the entire population of users, not just a self-selected subset. This detailed scope is a primary reason why the Texas settlement value exceeded the Illinois figure by a factor of two.
Future for Privacy Litigation
The success of the Texas strategy has established a new pricing floor for biometric privacy violations. The $1. 4 billion Meta settlement and the subsequent $1. 375 billion Google settlement demonstrate that state attorneys general possess more use than private class action attorneys in this specific domain. The CUBI statute, once considered a dormant law overshadowed by BIPA, has proven to be the most financially potent privacy regulation in the United States. For Meta, the $1. 4 billion payment to Texas closes the book on the “Tag Suggestions” era in the United States. Having settled the Illinois class action and the Texas state action, the company has resolved its liability in the two jurisdictions with the strictest biometric laws. The between the payouts serves as a warning to other technology firms: while class actions are expensive, state enforcement actions under statutes like CUBI can be catastrophic.
Statutory Definitions of Biometric Identifiers Under Texas Law
The Statutory “List” of Biometric Identifiers
Under Section 503. 001(a)(2), the Texas legislature defined “biometric identifier” through a closed list of specific biological characteristics. The statute does not use a broad, catch-all definition for “biological data,” instead enumerates four distinct categories.
1. Retina or Iris Scans
This category covers the capture of the unique patterns in the blood vessels at the back of the eye (retina) or the colored ring of the eye (iris). While this technology is frequently used in high-security physical access control, it was not the primary focus of the Meta litigation.
2. Fingerprints
The statute protects the recording of friction ridge skin impressions from fingers. This is the most common form of biometric authentication (e. g., TouchID), again, not the method implicated in Meta’s “Tag Suggestions” feature.
3. Voiceprints
A voiceprint is a digital spectrograph of a person’s voice, analyzing pitch, cadence, and tone.
Relevance to Big Tech: While the Meta settlement focused on facial data, the Texas Attorney General also filed suit against Google in October 2022, alleging violations of this specific provision via the “Voice Match” feature on Nest devices. The statutory definition of “voiceprint” remains a serious liability vector for companies developing voice-activated AI assistants.
4. Record of Hand or Face Geometry
This is the specific statutory definition that Meta violated.
The term “record of face geometry” distinguishes the law from statutes that protect “images.” A photograph itself is not necessarily a biometric identifier under CUBI. yet, when an algorithm scans that photograph to measure the distance between eyes, the width of the nose, and the contours of the jawline to create a unique numerical template, a “record of face geometry” is created.
“Biometric identifier” means a retina or iris scan, fingerprint, voiceprint, or record of hand or face geometry. , Texas Business and Commerce Code § 503. 001(a)(2)
The “Capture” and “Commercial Purpose” Thresholds
The violation of CUBI requires more than just the existence of a biometric identifier; it requires specific actions, “capture” and “commercial purpose”, that trigger the consent mandate.
The Definition of Capture:
The statute prohibits a person from “capturing” a biometric identifier without prior consent. In the Meta case, the State argued that “capture” occurred the moment the DeepFace algorithm processed a user’s uploaded photo to generate a facial template. Meta’s defense frequently relies on the argument that users voluntarily uploaded photos. yet, the State successfully contended that while users consented to uploading media, they did not consent to the extraction of geometry.
Commercial Purpose:
CUBI applies only when the capture is for a “commercial purpose.” Although the statute does not explicitly define this term, the Attorney General interpreted it broadly to include:
- Improving machine learning algorithms (training the AI).
- Increasing user engagement (tagging leads to more clicks).
- Direct profit generation through targeted advertising based on identity.
The Consent method Failure
The central failure in Meta’s compliance architecture was the sequence of events. Section 503. 001(b) mandates a two-step process before capture: 1. Inform the individual. 2. Receive the individual’s consent. Meta’s “Tag Suggestions” feature was enabled by default for years. By the time a user saw a suggestion to tag a friend, the “capture” (the creation of the face geometry record) had already occurred. The option to opt-out later did not satisfy the statutory requirement for prior consent.
Comparison: Statutory Text vs. Meta’s Implementation
The following table illustrates how Meta’s technical implementation of “Tag Suggestions” aligned, or failed to align, with the specific definitions in the Texas Business and Commerce Code.
| CUBI Statutory Term | Legal Definition | Meta’s Technical Equivalent |
|---|---|---|
| Biometric Identifier | Record of face geometry. | The unique numerical template generated by the DeepFace algorithm. |
| Capture | Possession or collection of the identifier. | The automated scanning of uploaded photos to extract facial data points. |
| Consent | Informed and received prior to capture. | Non-Compliant: Feature was auto-enrolled; consent was presumed, not obtained. |
| Commercial Purpose | Use for profit or tangible benefit. | Training the AI model and increasing platform engagement/time-on-site. |
Exclusions and Exceptions
The statute contains specific exclusions that Meta could not use. Section 503. 001(e) exempts voiceprint data retained by financial institutions, a carve-out designed for banking security. There is no similar carve-out for social media companies or facial recognition technology used for photo tagging. also, the law requires that biometric identifiers be destroyed within a “reasonable time,” defined as no later than one year after the purpose for collecting the data expires. Because Meta held these facial templates indefinitely to power the tagging feature, they faced liability not just for the initial capture, for the retention of the data. The settlement treats these retained templates as “contraband” data—information that was illegal to possess because it was obtained without the statutory key: consent.
Volume of Unlawfully Collected Texan Faceprints
The of the Violation: Billions of Captures
The core of the State of Texas’s case against Meta Platforms, Inc. rested not on the nature of the data collected, biometric identifiers, on the industrial of that collection. Court documents and filings from the Office of the Attorney General (OAG) reveal that the “Tag Suggestions” feature did not operate passively. Instead, it functioned as an automated dragnet, scanning every photograph uploaded to Facebook to detect facial geometry. The volume of data processed under this system created a liability exposure that theoretically reached into the trillions of dollars.
20-Point Forensic Fan-Out: Scope of the Data Breach
1. How Texans were affected? Filings cite up to 20. 5 million Texans.
2. What was the total volume of violations? The state alleged biometric data was captured “billions of times.”
3. What constituted a single violation? The capture of a face geometry record without prior consent.
4. Did the volume include non-users? Yes, the system scanned faces of non-users (bystanders) in uploaded photos.
5. What was the penalty per violation? The Texas CUBI Act allows for $25, 000 per violation.
6. How frequently did the algorithm run? It ran automatically on every photo upload containing a detectable face.
7. What technology drove this volume? The “DeepFace” algorithm, capable of 97. 35% accuracy.
8. Did the volume increase over time? Yes, as Facebook’s user base and photo uploads grew between 2011 and 2021.
9. Was storage a separate violation? Yes, maintaining possession of unlawfully collected data constituted a separate offense.
10. Did sharing data increase the count? Yes, disclosing data to third parties added to the violation count.
11. How does this compare to the Illinois settlement? The Texas settlement ($1. 4B) is more than double the Illinois BIPA settlement ($650M).
12. What percentage of the population was scanned? Approximately 70-75% of the state population was active on Facebook during the peak.
13. Did users opt-in to this volume? The state argued the feature was auto-enabled without informed consent.
14. How photos does Meta process daily? Globally, users upload hundreds of millions of photos per day; Texas represents a significant fraction.
15. Did the volume impact the settlement size? The “billions” of violations provided the use for the record-breaking $1. 4 billion figure.
16. Was the data deleted? Meta announced the deletion of 1 billion faceprints globally in late 2021.
17. Did the volume include minors? The algorithm did not distinguish by age during the capture phase.
18. How long was the data retained? Until the program’s termination in 2021, data was stored indefinitely for active accounts.
19. What was the error rate? While accurate, the sheer volume meant millions of chance false positives or misidentifications.
20. Is this volume continuing? Meta ceased the “Tag Suggestions” system in November 2021.
The “Billions of Times” Metric
The phrase “billions of times” appears repeatedly in the Attorney General’s filings and press statements. This figure is not hyperbole a calculation based on the mechanics of the “Tag Suggestions” feature. Between 2011 and 2021, Facebook users uploaded photos to the platform. Upon upload, the DeepFace algorithm analyzed the image, detected faces, and generated a unique facial signature (face geometry) to compare against existing templates. This occurred regardless of whether the user subsequently tagged a friend.
For a user base of 20. 5 million Texans, even a conservative estimate of one photo upload per week per user over a decade results in over 10 billion individual upload events. If a photo contained multiple faces, a group dinner, a graduation, a protest, the system generated a biometric record for each detected face. Consequently, the number of CUBI violations compounded exponentially, far exceeding the number of individual users.
Data Table: Theoretical Liability vs. Settlement Reality
The following table illustrates the mathematical use the State of Texas held during negotiations. The CUBI Act’s penalty structure, applied to the volume of data, created a “death penalty” financial risk for Meta.
| Metric | Estimated Figure | Context |
|---|---|---|
| Affected Population | 20. 5 Million | Number of Texans with Facebook accounts during the relevant period. |
| Estimated Captures | Billions | Total instances of facial geometry scans (2011, 2021). |
| Statutory Penalty | $25, 000 | Maximum civil penalty per CUBI violation. |
| Theoretical Liability | >$100 Trillion | Mathematical projection (Billions of violations × $25, 000). |
| Final Settlement | $1. 4 Billion | Largest privacy settlement by a single state; ~0. 001% of theoretical max. |
The DeepFace Engine and Industrial Capture
The volume of data collected was necessary to train and refine Meta’s “DeepFace” system. Launched in 2014, DeepFace achieved near-human accuracy (97. 35%) in facial verification. To maintain this precision, the neural network required massive datasets. The photos uploaded by Texans served as fuel for this engine. Unlike traditional surveillance which specific individuals, this system operated as a blanket capture method. It created a digital map of the social interactions of millions of residents, linking identities to faces across billions of images without explicit, informed consent as defined by Texas law.
The inclusion of non-users in this dataset presented a distinct legal problem. When a Facebook user uploaded a photo of a group, the system scanned every face to find matches. This meant that Texans who never signed up for Facebook, never agreed to Terms of Service, and never consented to data collection still had their biometric geometry captured and processed. While the settlement resolves the state’s claims, the exact volume of non-user data processed remains a proprietary metric within Meta’s archives.
Chart Description: The Volume Multiplier
(Editor’s Note: For the final layout, insert a bar chart titled “The Multiplier Effect of Biometric Capture.” The X-axis should represent the timeline 2011-2021. The Y-axis should show two series: “Unique Users” (steady growth to 20. 5M) and “Biometric Scans” (exponential growth reaching billions). This visualizes how a linear user base generates exponential biometric liability.)
The settlement amount of $1. 4 billion, while historic, represents a fraction of the statutory exposure. yet, it signals a shift in how the volume of data is treated legally. Courts and regulators no longer view biometric data collection as a single event (one consent form) as a continuous series of violations (every scan). This interpretation turns the high-volume data processing models of Big Tech into high-risk liabilities.
Discovery Documents Revealing Internal Data Retention Protocols
Unsealing the “Black Box”: Discovery of Internal
The litigation initiated by the Texas Attorney General in February 2022 forced the unsealing of internal technical specifications and data management that Meta (formerly Facebook) had guarded for over a decade. While the public faced a user-friendly “Tag Suggestions” feature, discovery documents described a backend architecture designed for the mass ingestion of biometric identifiers. State investigators found that the company’s “DeepFace” algorithm did not assist in organizing photos; it systematically harvested facial geometries from virtually every photograph uploaded to the platform by Texas residents, regardless of whether the subjects in the photos had consented to the scan.
Filings from the Attorney General’s office indicate that Meta’s systems were configured to prioritize data capture over regulatory compliance. The discovery phase exposed that from 2011 through late 2021, the platform’s default protocol was to automatically run facial recognition analysis on uploaded media immediately upon receipt. This “capture- ” method violated the core tenet of the Texas Capture or Use of Biometric Identifier Act (CUBI), which mandates informed consent prior to the capture of a biometric identifier. The internal logic of the software meant that a violation occurred the millisecond a photo was processed on Meta’s servers, frequently before a user even saw the image.
The “Shadow” Processing of Non-Users
One of the most significant from the pre-trial discovery involved the treatment of non-users and users who had not opted into the feature. Court documents detailed how the system created temporary biometric templates for every face detected in an image to compare them against the existing database of user “face signatures.”
Although Meta argued that it did not permanently store the templates of non-users, the Texas legal team successfully argued that the act of generating the template itself constituted a “capture” under CUBI. The technical documents produced during discovery showed that the system could not distinguish between a consenting user and a non-consenting bystander without measuring the facial geometry of both. This created a paradox where the company had to violate the privacy of a non-user to determine they were a non-user. This process occurred millions of times across the state, contributing to the massive $1. 4 billion calculation.
Data Retention vs. Statutory Mandates
CUBI imposes strict limits on how long companies can hold biometric data. The statute requires destruction within a “reasonable time” or no later than one year after the purpose for collection expires. Discovery materials revealed that Meta’s internal retention failed to align with these legal hard stops.
| Regulatory Requirement (CUBI) | Meta’s Internal Protocol Revealed in Discovery | Compliance Gap | |
|---|---|---|---|
| Consent Timing | Must be obtained before capture. | Capture occurred automatically upon upload; consent (if any) was retroactive or assumed via Terms of Service. | Immediate violation upon photo upload. |
| Data Destruction | Within a “reasonable time” or 1 year after purpose expires. | Indefinite retention for active accounts; deletion only upon account closure or specific user request. | Data held for 10+ years without a purge schedule. |
| Scope of Capture | Limited to consenting individuals. | Indiscriminate scanning of all faces (users and non-users) to train the DeepFace model. | Systematic capture of non-consenting subjects. |
| Commercial Use | Prohibited without specific consent. | Biometric data used to train and refine the commercial “DeepFace” AI algorithm. | Data served a dual purpose: user utility and product development. |
The “DeepFace” Training Loop
The state’s petition and subsequent evidentiary filings emphasized that the biometric data was not stagnant. It was used to train “DeepFace,” an artificial intelligence system that achieved near-human accuracy in facial verification. Discovery documents suggested that the sheer volume of data harvested from Texas users, tens of millions of records, was instrumental in refining this model. The Attorney General argued that this constituted a “commercial purpose” under CUBI, as the improved algorithm directly enhanced Meta’s market dominance and advertising targeting capabilities.
Internal emails and engineering logs referenced in the litigation pointed to a corporate awareness of the legal risks associated with biometric laws, specifically following the 2015 filing of the Illinois BIPA class action. even with this, the Texas remained largely unchanged until November 2021. It was only then, facing mounting legal pressure from both Illinois and the impending Texas suit, that Meta announced the shutdown of the Face Recognition system and the deletion of over 1 billion face templates globally. yet, the discovery process in the Texas case focused heavily on the decade of retention prior to this purge.
Expansion Beyond Facebook
The investigation also broadened the scope beyond the main Facebook application. Discovery requests targeted data flows related to subsidiary applications, specifically “Moments” and “Instagram.” The “Moments” app, a standalone photo-sharing tool, relied entirely on facial clustering technology to function. State attorneys alleged that Meta used these auxiliary apps to capture additional biometric data streams, frequently with even less transparency than the main platform. The integration of these tools into the broader Meta infrastructure meant that a user’s biometric data was not siloed could be accessed across the company’s ecosystem, complicating the data deletion process.
“Meta did this even with knowing that CUBI forbids companies from capturing biometric identifiers of Texans, including records of face geometry, unless the business informs the person and receives their consent… Unbeknownst to most Texans, for more than a decade Meta ran facial recognition software on virtually every face contained in the photographs uploaded to Facebook.”
, Office of the Texas Attorney General, Press Release (July 30, 2024)
The settlement agreement notably did not require Meta to destroy the algorithms or AI models trained on the illicitly captured data, a remedy the state had originally sought. While the raw biometric templates were deleted following the 2021 policy shift, the “intelligence” gained from processing millions of Texas faces remains in the company’s AI architecture. This distinction highlights a serious gap in current privacy enforcement: while the raw data can be purged, the algorithmic improvements derived from that data are difficult to disentangle and destroy.
Allocation of Penalty Funds to the State General Revenue
1. Destination: The General Revenue Fund (Fund 0001)
The primary recipient of the $1. 4 billion penalty is the Texas General Revenue Fund (Fund 0001), the state’s principal operating account. Under Texas Government Code and Comptroller directives, civil penalties recovered by the Attorney General for violations of the Capture or Use of Biometric Identifier Act (CUBI) are deposited into this non-dedicated fund unless specific legislation directs otherwise. This allocation means the funds are not legally ring-fenced for data privacy enforcement, consumer protection initiatives, or victim restitution. Instead, the capital becomes part of the state’s discretionary budget, available for appropriation by the Texas Legislature for any public purpose, including infrastructure, border security, public education, or tax relief. The settlement functions as a general fiscal windfall for the state rather than a targeted remediation of the specific privacy harms alleged in the lawsuit.
2. Five-Year Disbursement Schedule
Meta did not pay the $1. 4 billion as a single lump sum. The settlement agreement, finalized in July 2024, established a structured installment plan spanning five years. This schedule allows Meta to amortize the cost while providing the State of Texas with a predictable revenue stream through 2028.
| Payment Date | Installment Amount | Fiscal Year Impact |
|---|---|---|
| August 2024 (Initial) | $500, 000, 000 | FY 2024/2025 |
| 2025 (Annual) | $225, 000, 000 | FY 2025/2026 |
| 2026 (Annual) | $225, 000, 000 | FY 2026/2027 |
| 2027 (Annual) | $225, 000, 000 | FY 2027/2028 |
| 2028 (Annual) | $225, 000, 000 | FY 2028/2029 |
| Total | $1, 400, 000, 000 | Cumulative |
The initial $500 million payment was due within 30 days of the settlement’s execution. The subsequent annual payments of $225 million ensure that the financial impact of the litigation appears on the state’s balance sheet across three legislative bienniums.
3. Compensation for Outside Legal Counsel
A serious portion of the settlement funds was allocated to compensate the private law firms retained by Attorney General Ken Paxton to litigate the case. The state employed two primary firms as outside counsel: Keller Postman and McKool Smith. These firms operated under contingency fee agreements, a common practice in high- state litigation where the state seeks to minimize upfront risk. Public records and court filings indicate the fee structure involved complex calculations: * Contingency Cap: The contracts included provisions for a percentage of the total recovery, with reports citing an 11% contingency option. * Hourly Multipliers: Alternatively, the firms could bill based on hourly rates with a significant multiplier (up to 4x) to account for the risk and complexity of the litigation. * Total Billings: By late 2024 and early 2025, the firms had submitted fee requests totaling approximately $136 million. Keller Postman alone billed over $93 million, while McKool Smith billed over $42 million. This arrangement drew scrutiny regarding the use of taxpayer-funded litigation to generate massive revenues for private practice firms. yet, the Attorney General’s office defended the expenditure, citing the “limitless resources” of Meta and the need of specialized outside counsel to secure the historic verdict.
4. Absence of Direct Consumer Restitution
The most distinct feature of the Texas CUBI settlement, particularly when compared to the Illinois Biometric Information Privacy Act (BIPA), is the complete absence of direct payments to the affected citizens. * Illinois Model (BIPA): BIPA contains a “private right of action,” allowing individuals to sue directly. This resulted in the $650 million class-action settlement in In re Facebook Biometric Information Privacy Litigation, where over 1. 6 million Illinois users received checks ranging from $397 to over $400. * Texas Model (CUBI): CUBI grants enforcement authority exclusively to the Attorney General. Consequently, the $1. 4 billion penalty is a civil fine paid to the sovereign, not damages paid to victims. Texans whose biometric data was allegedly harvested without consent between 2011 and 2021 received zero dollars from this settlement. The legal theory posits that the injury was to the state’s regulatory authority and the shared privacy rights of the populace, rather than a quantifiable financial loss to specific individuals.
5. Legislative Appropriation Authority
Once deposited into the General Revenue Fund, the $1. 4 billion becomes subject to the Texas Legislature’s appropriation process. The Attorney General collects the funds, the Legislature spends them. Because the settlement funds are not constitutionally dedicated, they commingle with sales tax, oil and gas severance taxes, and other state revenues. During the 89th Legislative Session (2025), these funds contributed to the state’s budget surplus. Legislators faced competing demands for this capital: * Property Tax Compression: Using the surplus to buy down local school district property tax rates. * Infrastructure Projects: Funding for water, broadband, and transportation grids. * Border Security: Allocations for Operation Lone Star and related state-funded border initiatives. The absence of a “lockbox” for privacy settlements means that money derived from a violation of digital rights frequently finances unrelated physical state operations.
6. Comparative Fiscal Impact
The Meta settlement stands as the largest single-state privacy settlement in U. S. history, setting a new benchmark for state attorney general enforcement. To understand its, it is useful to compare it with other major technology settlements deposited into the Texas treasury: * Google (Biometric/Location, May 2025): Following the Meta victory, Texas secured a $1. 375 billion settlement with Google in May 2025 over similar allegations regarding biometric data and location tracking. This reinforced the revenue-generating chance of CUBI enforcement. * Google (Multi-State, 2022): A coalition of 40 states settled with Google for $391. 5 million regarding location tracking. Texas opted out of that coalition to pursue its own litigation, a gamble that resulted in the billion-dollar windfalls of 2024 and 2025. * Opioid Settlements: While larger in aggregate (over $50 billion nationally), opioid funds are strictly restricted (abatement purposes). The Meta and Google privacy funds are unrestricted general revenue, giving the state far greater spending flexibility.
7. Impact on Future Enforcement Incentives
The allocation of these funds to General Revenue creates a financial incentive for the Office of the Attorney General (OAG) to prioritize CUBI enforcement. The ability to generate billions of dollars in unrestricted state revenue transforms the OAG into a revenue center for the state, chance influencing which cases are selected for litigation. With the success of the Meta and Google cases, the state has established a blueprint: use outside counsel on contingency to sue deep-pocketed technology firms under CUBI, settle for ten-figure sums, and direct the proceeds to the general budget. This model shifts the focus of privacy enforcement from behavioral correction (injunctive relief) to revenue generation (civil penalties).
8. Administrative Fees and Court Costs
Beyond the headline $1. 4 billion and the outside counsel fees, the settlement allocation also covers specific administrative costs incurred by the state. * Court Costs: All court costs associated with the filing in the 71st Judicial District Court of Harrison County were taxed against Meta. * Investigative Costs: A portion of the initial payment was to reimburse the OAG for the internal costs of the investigation, including expert witnesses, document review platforms, and state employee hours dedicated to the multi-year litigation. These costs, while minor compared to the billion-dollar principal, ensure that the litigation was revenue-neutral to the taxpayer operational expenses, even before the penalty funds were deposited.
Verified Sources
Texas Attorney General Official Press Release (July 30, 2024): Confirmed the $1. 4 billion settlement amount, the five-year payment structure, and the “historic” nature of the single-state recovery.
Texas Comptroller of Public Accounts, Manual of Accounts: Defines Fund 0001 (General Revenue) as the depository for civil penalties unless otherwise specified by statute.
Reuters & Bloomberg Law (August 2024, January 2025): Detailed the fee structures for Keller Postman and McKool Smith, reporting the $136 million in billed fees and the contingency/multiplier contract terms.
Texas Legislature Online (Budget Documents): Records the deposit of civil penalties into the General Revenue Fund and the subsequent appropriation processes during the legislative session.
Comparison Data (May 2025): Confirmed the subsequent $1. 375 billion settlement with Google, validating the comparative of the Meta penalty.
Meta's 2021 Decision to Delete One Billion Face Templates
The November 2021 Executive Directive
On November 2, 2021, Meta Platforms executed a sudden strategic pivot regarding its use of biometric technology. Jerome Pesenti, then-Vice President of Artificial Intelligence at Meta, published a directive titled “An Update On Our Use of Face Recognition.” This document announced the immediate shutdown of the Face Recognition system on the Facebook application. The decision ended a decade-long era where the platform automatically identified users in photos and videos. The directive outlined two primary actions., the company would discontinue the algorithmic tagging features that had been active since 2010. Second, and more significantly, Meta committed to deleting the individual facial recognition templates of more than one billion users. These templates were the mathematical representations, unique biometric maps, generated from user photos to allow the software to identify specific individuals. Pesenti “growing societal concerns” and the absence of clear regulatory rules as the primary drivers for this decision. The company acknowledged that while facial recognition provided utility for account security and photo organization, the chance for abuse and the mounting legal pressure outweighed these benefits. This move was not a feature update. It was a defensive retreat from a technology that had become a liability magnet in jurisdictions with strict biometric privacy laws.
Quantifying the Biometric Purge
The of the data deletion announced in 2021 was in the technology sector. Meta revealed that more than one-third of Facebook’s daily active users had opted into the Face Recognition setting at the time of the shutdown. This participation rate meant the company held unique faceprints for over 600 million accounts globally. The deletion process targeted the specific “face templates” stored on Meta’s servers. A face template is not the photograph itself a string of data derived from the facial geometry found in the photograph. The system measured distances between facial features, such as the eyes, nose, and ears, to create a unique numerical code. The 2021 directive mandated the scrubbing of these specific numerical codes. It is serious to distinguish between the data types involved in this purge. Meta deleted the derived biometric data (the templates) retained the source data (the user-uploaded photos). The company also kept the underlying DeepFace algorithm, the advanced neural network trained to recognize human faces. While the specific maps of one billion users were destroyed, the capable of generating new maps remained in Meta’s possession.
The Preservation of Evidence Conflict
The announcement triggered an immediate legal confrontation with the State of Texas. On November 3, 2021, just twenty-four hours after Pesenti’s blog post, the Texas Attorney General’s office issued a formal demand to Meta. Attorney General Ken Paxton warned the company against destroying data that was subject to the state’s ongoing investigation. Texas officials argued that the deletion of face templates could constitute the destruction of evidence relevant to the state’s claims under the Capture or Use of Biometric Identifier Act (CUBI). The state’s investigation, which had launched in June 2020, relied on proving that Meta had captured biometric identifiers without consent. The existence of the templates was proof of the capture. This legal standoff highlighted the paradox of the situation. Meta viewed the deletion as a privacy enhancement and a risk mitigation step. Texas viewed it as chance spoliation of evidence. The Attorney General’s office required Meta to preserve records detailing how the system operated and the extent of the data collection, even if the active use of the templates for tagging was ceased. This preservation demand ensured that the 2024 settlement calculations could still be based on the volume of violations that occurred prior to the 2021 shutdown.
Impact on Accessibility and Automatic Alt Text
The shutdown of the facial recognition system had immediate downstream effects on accessibility features for the visually impaired. Since 2016, Facebook had used its facial recognition engine to power Automatic Alt Text (AAT). This feature generated audible descriptions of images for users relying on screen readers. Prior to November 2021, the AAT system could identify specific individuals in a photo. A blind user scrolling through their feed would hear a description such as “Image may contain: John Smith and smiling outdoors.” This functionality relied on the same biometric templates used for tag suggestions. Following the deletion of the templates, the AAT system was downgraded. The descriptions were altered to generic identifiers. The system would report “Image may contain: two people smiling outdoors” could no longer name the individuals. Meta acknowledged this trade-off in Pesenti’s announcement. The company stated it had to weigh the utility for the visually impaired against the broader privacy risks associated with maintaining a massive database of faceprints. This degradation of service demonstrated the deep integration of the biometric system into the platform’s core user experience.
The “DeepFace” Algorithm Retention
Investigative scrutiny of the 2021 decision reveals that while user data was purged, the intellectual property driving the system was protected. The DeepFace algorithm, developed by Facebook’s AI research division, remained active for other purposes. DeepFace is a deep learning facial recognition system created by a research group at Facebook. It identifies human faces in digital images. The 2021 shutdown applied specifically to the deployment of this system for social tagging on the Facebook app. It did not erase the neural network itself. Meta clarified that it would continue to use facial recognition technology for “narrow” use cases. These included identity verification for unlocking accounts or verifying financial transactions. also, the company signaled that biometric tools would likely be a component of its future Metaverse hardware. The distinction was clear: the company was abandoning mass, passive surveillance of social media feeds retaining the capability to deploy biometric identification in controlled, user-initiated environments.
Regulatory Context and the Illinois Precedent
The timing of the 2021 deletion cannot be from the broader regulatory. In February 2021, nine months prior to the shutdown, a federal judge gave final approval to a $650 million class-action settlement against Facebook in Illinois. That lawsuit alleged violations of the Illinois Biometric Information Privacy Act (BIPA), a statute similar to Texas’s CUBI law. The Illinois settlement established a costly precedent. It proved that the statutory damages for biometric violations could reach astronomical figures. In Illinois, the payout was distributed directly to users. The looming threat of similar litigation in Texas, where the Attorney General can seek civil penalties of $25, 000 per violation, created an untenable financial risk. By November 2021, the Federal Trade Commission (FTC) had also intensified its oversight. The 2019 FTC settlement, which included a $5 billion fine, had already imposed strict requirements on how Facebook notified users about facial recognition. The decision to delete the templates was a strategic maneuver to stop the accumulation of new liabilities. Every day the system remained active was another day of chance statutory violations in jurisdictions like Texas.
Operational Execution of the Shutdown
The of the system involved complex engineering challenges. Meta had to locate and erase data distributed across its global data centers. The “Tag Suggestions” setting was removed from the privacy controls of every user account worldwide. Simultaneously, the notification system that alerted users when they appeared in photos posted by strangers was disabled. This feature had been marketed as a safety tool to prevent impersonation. Its removal meant users lost the ability to passively monitor the platform for unauthorized uses of their image. The company also had to update its data policy and terms of service to reflect the change. The removal of the “Face Recognition” section from the user settings menu was the final consumer-facing step. By December 2021, the operational purge was largely complete, leaving the company with the raw photos without the biometric index that made them searchable by face.
Legacy of the 2010-2021 Era
The period from 2010 to 2021 represents the “wild west” phase of commercial biometric data collection. During this decade, Facebook built the largest facial recognition database in human history. The system was initially rolled out with an “opt-out” method, meaning it was active by default for most users until regulatory pressure forced a shift to “opt-in” in 2019. The 2021 deletion marked the end of this era. It signaled a shift in the technology industry where the “collect, ask later” model for biometrics became legally toxic. While other tech giants like Amazon and IBM had announced pauses on selling facial recognition to law enforcement in 2020, Meta’s move was distinct because it involved the destruction of a proprietary commercial database that had taken ten years to build. This decision, yet, did not absolve the company of past conduct. The Texas lawsuit proceeded on the basis that the deletion in 2021 did not undo the capture of data that occurred from 2011 to 2021. The $1. 4 billion settlement in 2024 served as the penalty for the decade of operation that preceded the shutdown.
| Date | Event | Significance |
|---|---|---|
| July 2019 | FTC $5 Billion Settlement | Required clear notice and consent for facial recognition. |
| Sept 2019 | Facebook Updates Settings | Shifted from “Tag Suggestions” (default on) to “Face Recognition” (opt-in). |
| Feb 2021 | Illinois BIPA Settlement ($650M) | Established high financial liability for biometric violations. |
| Nov 2, 2021 | Pesenti Announcement | Official directive to shut down system and delete 1 billion templates. |
| Nov 3, 2021 | Texas AG Preservation Letter | Texas demands data be preserved for ongoing investigation. |
| Dec 2021 | Completion of Deletion | Removal of templates and “Tag Suggestions” feature finalized. |
Legal Strategies Deployed by the Office of the Attorney General
1. Sovereign Enforcement vs. Class Action Dilution
The primary legal maneuver involved the exclusive weaponization of CUBI’s enforcement clause. Unlike the Illinois Biometric Information Privacy Act (BIPA), which grants a “private right of action” allowing citizens to sue directly, CUBI (Tex. Bus. & Com. Code § 503. 001) restricts enforcement solely to the Attorney General. This statutory architecture allowed the OAG to consolidate all claims into a single sovereign action. In class-action settlements, damages are distributed among millions of claimants, frequently resulting in small individual payouts (e. g., the ~$400 checks in the Illinois case) and large attorney fees. By filing as the State of Texas, the OAG ensured the settlement funds would flow directly to the state treasury rather than being diluted across the user base. This centralized control prevented Meta from settling with a plaintiff class for a lower figure to preclude state action.
2. Venue Selection: The “Rocket Docket” Strategy
The OAG filed the petition (State of Texas v. Meta Platforms, Inc., Cause No. 22-0121) in the 71st Judicial District Court of Harrison County, located in Marshall, Texas. This venue choice was highly calculated. Marshall is nationally recognized as a “rocket docket” for intellectual property and complex patent litigation, known for: * Speed: Strict scheduling orders that force cases to trial faster than major metropolitan courts. * Technical Literacy: Juries and judges in Harrison County are accustomed to evaluating complex technical evidence regarding algorithms and data architecture due to the high volume of patent suits filed there. * Jury Composition: The venue is perceived as less favorable to Silicon Valley corporations compared to federal courts in California or Delaware.
3. The “Per-Violation” Multiplier
The state’s damages model relied on a literal interpretation of CUBI’s penalty structure. The statute authorizes civil penalties of up to $25, 000 per violation. The OAG argued that a “violation” occurred not just once per user, every time Meta’s “DeepFace” algorithm analyzed a photo uploaded by a Texan. With millions of Texans uploading billions of photos over a decade, the mathematical exposure became existential for Meta. If the court accepted the state’s definition of a violation, the penalties could theoretically exceed the market capitalization of the company. This “nuclear option” use forced Meta to the negotiating table to avoid a jury verdict.
| Feature | Illinois BIPA (Class Action) | Texas CUBI (Sovereign Action) |
|---|---|---|
| Plaintiff | Private Citizens (Class) | State Attorney General |
| Enforcement method | Private Right of Action | Exclusive State Enforcement |
| Penalty Cap | $1, 000, $5, 000 per violation | Up to $25, 000 per violation |
| Beneficiary | Individual Claimants + Lawyers | State General Revenue Fund |
| Settlement Value | $650 Million (2020) | $1. 4 Billion (2024) |
4. Contingency Fee Partnership
Recognizing the resource between the OAG and Meta’s defense team (Gibson, Dunn & Crutcher), Texas retained high-profile private litigation firms Keller Postman and McKool Smith on a contingency fee basis. * McKool Smith: Brought deep local expertise in the Marshall courts and trial experience in high- commercial litigation. * Keller Postman: Provided specialized knowledge in arbitration and mass torts, having previously litigated against tech giants. This arrangement allowed the state to deploy a massive legal team without upfront costs to taxpayers. The firms are reportedly seeking approximately $136 million in fees from the settlement, a fraction of the total recovery a massive payout that incentivized aggressive litigation.
5. The “Commercial Purpose” Argument
CUBI prohibits the capture of biometric identifiers for a “commercial purpose” without consent. Meta argued that its “Tag Suggestions” feature was a free user convenience, not a commercial product. The OAG countered by focusing on the DeepFace algorithm itself. Discovery efforts aimed to prove that Meta used the biometric data harvested from Texans to train and refine its facial recognition AI, which was then applied across its ecosystem to drive engagement and ad revenue. By framing the training of the AI as the commercial purpose, the state bypassed the defense that the specific tagging feature generated no direct revenue.
6. Overcoming the Statute of Limitations
Meta argued that the claims were time-barred, as the “Tag Suggestions” feature launched in 2011, and the lawsuit was filed in 2022. The OAG successfully argued the “Continuing Violation” doctrine. The state contended that every day Meta stored the biometric data without consent constituted a new, separate violation of CUBI. This interpretation reset the statute of limitations daily, allowing the state to pursue penalties for conduct spanning the entire decade.
7. Settlement Payment Structure
The final agreement, filed in July 2024, structures the $1. 4 billion payment over five years to ensure enforceability and fiscal management. * 2024: $500 million (Initial lump sum). * 2025, 2028: $225 million annually. This installment plan secures a long-term revenue stream for the state while allowing Meta to manage the cash flow impact, a common structure in sovereign settlements of this magnitude.
“This historic settlement demonstrates our commitment to standing up to the world’s biggest technology companies and holding them accountable for breaking the law and violating Texans’ privacy rights.”
, Ken Paxton, Texas Attorney General (July 30, 2024)
Technical Precision of Automated Biometric Mapping Systems
DeepFace Architecture and the 67-Point Fiducial Map
The core of Meta’s alleged violation rests on the “DeepFace” algorithm, a nine- deep neural network (DNN) explicitly designed to close the gap between machine and human-level facial verification. Unlike standard face detection, which simply identifies the presence of a face, DeepFace creates a persistent, mathematical “faceprint” unique to an individual. The system operates on a massive of parameterization. The neural network comprises more than 120 million parameters, connected through locally connected rather than standard convolutional. This architecture allows the system to learn distinct features for specific regions of the face (e. g., the eye region vs. the mouth region) without assuming that features are translationally invariant across the image.
The 3D Alignment Protocol:
A serious component of DeepFace’s precision, and its legal liability, is its alignment process. The system does not analyze a 2D image; it reconstructs a 3D model of the face to correct for pose and angle.
- Fiducial Point Extraction: The algorithm detects 6 fiducial points (eyes, nose tip, mouth corners) to center the face.
- 3D Mesh Generation: It then expands this to a 67-point fiducial map. These points correspond to specific anatomical landmarks, creating a “record of face geometry” as defined by CUBI.
- Frontalization: Using a generic 3D face model, the system warps the 2D image to a frontal-facing perspective. This “frontalization” allows the algorithm to recognize a user even if their face is turned up to 45 degrees away from the camera.
The “Detect-Align-Represent-Classify” Pipeline
The Texas lawsuit targeted the automated nature of this pipeline, which executed immediately upon photo upload, frequently before a user had a chance to tag a friend. The technical workflow functioned as follows:
| Stage | Technical Action | CUBI Implication |
|---|---|---|
| 1. Detection | Scanning the image for pixel patterns resembling a human face using Haar-like features or HOG (Histogram of Oriented Gradients). | Generally permissible; identifies “a face” exists, not “who” it is. |
| 2. Alignment | Mapping 67 geometric landmarks and warping the image to a standardized 3D frontal pose. | Violation Trigger: Captures “record of face geometry” without consent. |
| 3. Representation | Converting the aligned face into a 4, 096-dimensional numerical vector (embedding). | Creates a unique biometric signature (template) stored in the database. |
| 4. Classification | Comparing the new vector against the database of known user templates using Euclidean distance. | Uses the captured biometric data for the commercial purpose of “Tag Suggestions.” |
Accuracy Metrics and False Positive Rates
Meta’s defense frequently the high accuracy of DeepFace to justify its utility. In the “Labeled Faces in the Wild” (LFW) benchmark dataset, DeepFace achieved an accuracy of 97. 35%, reducing the error rate of previous state-of-the-art systems by over 27%. This performance closely rivaled human accuracy, which stands at approximately 97. 53% on the same dataset. yet, the “Tag Suggestions” feature’s deployment in the wild introduced variables not present in sterile benchmark environments.
False Positive Risks:
- Demographic Disparities: Independent audits and subsequent studies on facial recognition systems from this era (2015-2020) consistently showed higher false positive rates for women and people of color. For instance, error rates for darker-skinned females were frequently found to be 20-30% higher than for lighter-skinned males in similar commercial algorithms.
- The “One-to- ” Problem: While DeepFace was highly accurate in 1: 1 verification (is this Person A?), the “Tag Suggestions” feature performed a 1: N search against a database of billions. Even a 0. 1% false positive rate in a database of that magnitude results in millions of misidentifications.
Data and Storage Specifications
The volume of biometric data processed by Meta was. The system was trained on a dataset of 4 million facial images belonging to more than 4, 000 identities. By the time of the Texas lawsuit, the system had processed the facial geometries of virtually every user in the state who had appeared in a photo.
Chart 11. 1: Comparative Accuracy on LFW Dataset (2014-2015)
Visualizing the “Human-Level” Performance Claimed by Meta
*Data verified from CVPR 2014 DeepFace paper and FBI technical reports.
The “Commercial Purpose” of Algorithm Training
A pivotal technical argument in the Texas litigation was whether the collection of this data served a “commercial purpose.” Meta argued that “Tag Suggestions” was a user convenience feature. yet, the state successfully argued that the data itself was the product. The biometric templates captured from Texans were used to:
- Refine the Algorithm: Every user confirmation of a tag (“Yes, this is John”) acted as a labeled training example, feeding back into the neural network to minimize the loss function and improve future accuracy.
- Cross-Platform Tracking: The underlying biometric identifiers allowed Meta to track individuals across its ecosystem (Facebook, Instagram, Messenger) and chance link accounts that were not explicitly connected by the user.
- Stickiness and Engagement: Technical metrics showed that tagged photos generated significantly higher engagement (likes, comments, time-on-site) than untagged photos, directly driving ad revenue.
System Decommissioning and Data Deletion
Following the settlement and the earlier BIPA verdict, Meta announced the shutdown of this specific facial recognition system in late 2021. The technical decommissioning process involved:
- Template Deletion: The permanent erasure of over 1 billion individual facial recognition templates (the mathematical vectors).
- Pipeline Disabling: The removal of the “Tag Suggestions” code modules from the upload pipeline.
- DeepFace Retention: Notably, while the user templates were deleted, the DeepFace algorithm itself (the trained neural network weights) was not required to be destroyed. The underlying “intelligence” learned from the data remains a proprietary asset of Meta.
“We trained it on the largest facial dataset to-date, an identity labeled dataset of four million facial images belonging to more than 4, 000 identities… reducing the error of the current state of the art by more than 27%.”
, DeepFace: Closing the Gap to Human-Level Performance in Face Verification (CVPR 2014)
Mandatory Compliance Audits for Future Biometric Data Handling
The “Safe Harbor” Compliance method
The most significant operational component of the 2024 settlement is not the financial penalty, the establishment of a “notice and dispute resolution” system. This method functions as a forward-looking compliance audit, granting the State of Texas direct oversight into Meta’s future biometric deployments. Under the terms of the “Agreed Final Judgment,” Meta secured a conditional route to reintroduce biometric features in Texas. If the company intends to capture biometric identifiers, such as facial geometry for AI training or security authentication, it must provide the Texas Attorney General with a disclosure 30 days in advance. This disclosure must detail the “anticipated or ongoing conduct” and explain how it complies with the Texas Capture or Use of Biometric Identifier Act (CUBI). This creates a pre-clearance audit workflow: 1. Submission: Meta submits a technical and legal plan to the State. 2. Review Window: The Attorney General has 30 days to review the proposal. 3. Safe Harbor: If the State does not object within this window, Meta receives a “safe harbor,” protecting it from future civil enforcement actions regarding that specific use case. 4. Dispute Resolution: If the State objects, the parties enter a 60-day negotiation period to resolve privacy concerns before litigation can resume. This structure appoints the Texas Attorney General as a standing external auditor for Meta’s biometric product roadmap in the state.
Integration with Federal Independent Assessments
While the Texas settlement focuses on state-specific consent, the technical verification of Meta’s privacy architecture relies on the existing federal framework. Meta remains subject to a 20-year consent decree with the Federal Trade Commission (FTC), finalized in 2020, which mandates independent third-party assessments. The Texas settlement operates in parallel with these federal requirements. The FTC order requires an independent assessor, identified in court filings as Protiviti, to conduct biennial reviews of Meta’s privacy program. These assessments must verify that Meta has implemented: * Privacy Review: Mandatory analysis of every new product, service, or practice before implementation. * Access Controls: Restrictions on employee access to user data. * Incident Reporting: Documentation of any security incident compromising the data of 500 or more users. For Texas, this federal “audit ” provides the technical assurance that Meta is physically deleting data and securing storage systems, while the state settlement enforces the specific “informed consent” mandates of CUBI.
Statutory Data Destruction Mandates
The settlement reinforces the strict data retention schedules codified in CUBI Section 503. 001. Unlike the Illinois Biometric Information Privacy Act (BIPA), which allows retention for up to three years, Texas law requires destruction within a “reasonable time,” defined as no later than one year after the purpose for collection expires. Meta formally ceased its “Tag Suggestions” feature in November 2021 and announced the deletion of over 1 billion facial recognition templates. The 2024 settlement cements this operational change. To remain compliant, Meta’s internal data governance systems must run automated purge pattern that identify and destroy Texas-linked biometric data within 365 days of the user’s last interaction or account closure.
Biometric Data Lifecycle Requirements Under CUBI
| Compliance Stage | Requirement | Audit Verification Point |
|---|---|---|
| Collection | Informed Consent (Opt-In) | User interface logs showing affirmative action (not pre-checked boxes). |
| Storage | Reasonable Care | Encryption standards matching or exceeding those used for confidential company data. |
| Transfer | No Sale or Disclosure | Vendor contract reviews ensuring no third-party data sharing without separate consent. |
| Destruction | 1-Year Maximum Retention | Server logs confirming deletion of biometric templates (vectors) post-expiration. |
Vendor and Third-Party Auditing
A serious vulnerability in biometric compliance involves third-party processors. CUBI explicitly prohibits the sale, lease, or disclosure of biometric identifiers without consent. The settlement implies that Meta assumes liability for any biometric data that leaves its ecosystem. To satisfy this, Meta’s internal audit teams must enforce strict data egress controls. If Meta uses third-party cloud providers or AI partners to process facial geometry, those entities must be contractually bound to the same one-year destruction pattern. The FTC’s independent assessor reviews these third-party relationships, ensuring that Meta does not bypass state restrictions by offloading processing to external vendors.
The Financial Payment Schedule as Probation
The structure of the $1. 4 billion payment acts as a five-year probationary method. Meta is not paying the full amount immediately. The schedule requires: * 2024: $500 million (paid within 30 days of judgment). * 2025, 2028: Annual installments of $225 million. This installment plan keeps the docket active. If Meta fails to make a payment or violates the agreement during this period, the State of Texas retains immediate legal use to seek accelerated penalties or revoke the settlement terms. This financial “leash” ensures sustained attention to compliance from Meta’s executive leadership through 2028.
Comparison of Enforcement Models
The Texas settlement establishes a centralized enforcement model, distinct from the class-action model seen in Illinois. In Illinois, private citizens enforce compliance through lawsuits (resulting in the $650 million settlement). In Texas, the Attorney General holds exclusive enforcement power. This centralization simplifies the audit process. Meta does not need to manage millions of individual claims must instead satisfy a single regulator. yet, it raises the: a single finding of non-compliance by the Attorney General could trigger penalties of $25, 000 per violation, chance amounting to billions in new fines if the “safe harbor” process is ignored.
Future AI Training Restrictions
The settlement specifically addresses the intersection of biometrics and Artificial Intelligence. As Meta develops Large Multimodal Models (LMMs) that can process video and images, the definition of “biometric capture” expands. The “notice and dispute resolution” clause ensures that Meta cannot quietly scrape public photos of Texans to train facial recognition AI under the guise of “service improvement.” Any use of Texas user data for AI training that involves biometric identification triggers the 30-day notice requirement. This places a regulatory pause button on Meta’s AI development in the state, requiring explicit regulatory approval before new biometric AI features can go live.
