Checkr's automated background-check platform

What this models

This example runs on the Screening-vendor-class gig background reporting network: 12 components and 25 pathways between them. Every context in the Lab is a stylized model, never a reconstruction of any actual deployment, and each assumption behind it carries a provenance label.

Evidence base: 3 assumed · 20 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

• assumed

  This models the shared-screening-vendor pattern documented in the Checkr gig background screening case file — the class of deployment where the failing component sits at a consumer reporting agency one hop upstream of every party that acts on its output. It is not a reconstruction of any vendor's code, and no component here is a learned model: the record describes identifier matching, charge normalisation and rule evaluation against a customer's criteria.

• baseline

  LEG RE-DERIVATION (2026-09-22, car LEG-3). This board was re-drawn at the coarsest granularity that still distinguishes every documented mechanism of the deployment, and it lands at twelve nodes and twenty-five pathways. One node and nine pathways were folded into survivors that already carry the same documented flow, and every folded fact is stated on its survivor. (1) The ongoing record-event feed behind the standing-monitoring product is carried on the furnished file and on the write that creates it, because what that supply does is re-open a file that already travels. (2) The agency's own exception path is drawn as one read and one write rather than four pathways: the match exceptions that reach those staff, and the same review reaching back into the source aggregation, are stated on the read. (3) The furnished adjudication reaching the correction channel is stated on that channel's read of the disputed file, with the timing it carried — the channel opens on the subject's initiative, once the platform has already decided. (4) The stated standard reaching the matching procedure is carried on the regulator's own check of that procedure, which already states the same standard and the same withdrawal. (5) The platform working the furnished file, and recording its access decision on its own side, are stated on the pathway that furnishes the report. (6) A file entering the correction record when somebody disputes it is stated on the write that fills that record. (7) The aggregation reaching a person's file is stated on the write the match makes, which is the route the record itself describes. Nothing this file said about the deployment before is unsaid now; the same deployment is drawn with fewer separate elements.

• baseline

  Two model nodes are drawn because the record documents two components whose errors are different in kind. The matcher's error is a record attached to the wrong person, which is what the January 2026 nationwide class pleads. The adjudication layer's error is somebody else's rule applied to facts that may be correct, which is what the 2021 pleading describes: a 2013 Virginia speeding offence that genuinely was a misdemeanour where it happened and a civil infraction where the work was, adjudicated without the individualised fair-chance analysis the city required. Merging them into one component would merge two pleadings the dossier keeps apart.

• baseline

  Two input feeds are drawn because two separate supplies are documented with two different owners, and the third supply this record names is carried on the store it feeds. The county, state and national record sources are external and their completeness and disposition quality vary, which is the January 2024 advisory opinion's whole subject. The identity graph is the vendor's own asset, and its scale is a vendor claim of 260 million identities covering 96 percent of United States adults. The third is the ongoing record-event feed behind the standing-monitoring product, and it is documented: new criminal-record and motor-vehicle events arriving for someone who was screened once and has been working since, which make a person who applied once re-screenable without applying again. What that supply does is re-open a file that already travels, so it is stated on the furnished file and on the write that creates it, which is also how the source model this network is derived from carries it. Motor-vehicle records are folded into the first feed rather than drawn separately, because the record documents them as a record type rather than as a separate pathway.

• baseline

  The aggregated criminal-record store is drawn separately from the external court sources because the gap between them is the regulator's own stated concern. The pipeline reads a pool assembled ahead of any particular request, which is what an instant check returning in under a second requires, and the January 2024 advisory opinion's four duties — suppressing expunged, sealed and legally restricted items, carrying dispositions with reported arrests and charges, eliminating duplicates, honouring reporting windows — are all duties about that pool. Those are findings about the deployment class and about a neighbouring screening market, and this file attributes none of them to this vendor.

• baseline

  The furnished consumer report file is the product, and the fan-out lives there rather than in any single pathway. One file is furnished to the platform that ordered the check, re-furnished on later checks by any platform sharing the vendor — one platform's own policy re-checks incumbent drivers at least every two years — and re-opened by the continuous-monitoring product when a new record event arrives. No source documents a specific multi-platform simultaneous adverse action; the fan-out is entailed by the shared-vendor topology and the vendor's own market-share claim, and this file narrates it in exactly that register.

• baseline

  The self-loop on the matcher reads 3 on the vendor's own marketing claim that ninety percent of gig-economy background checks run through this one pipeline, alongside 140,000-plus business customers. It is the mechanism the volume actually creates: a matching rule's defect repeats identically across every platform that buys from it rather than varying vendor by vendor. Every figure in that sentence is the vendor's, fetched from its own pages on 28 August 2026, and self-serving in both directions — it is capability marketing and it is also an admission of concentration.

• baseline

  Two operator classes are drawn and the screened worker is neither of them. The worker is the SUBJECT of the report, which is a served-person position outside these dynamics; the worker's dispute enters this network as the signal that opens the statutory correction channel, drawn as the rung on that channel rather than as a class. Modelling the worker as an operator would put a served person inside the dynamics, which this catalogue does not do. The consequence is stated rather than drawn: the pathway from an adverse action to a dispute runs through a person this diagram does not carry, and the 2021 pleading documents that adverse action arriving with no notice, no process and no communication, so the worker had to discover the cause unaided.

• baseline

  Three reviewer nodes are drawn because three correction channels are documented, and each is wired in by its own read of different content. The statutory reinvestigation channel reads the disputed file itself and the procedure it must follow. The regulators read the complaint and guidance record they maintain. The private enforcement channel reads the furnished files two class complaints were built on. None of the three dangles.

• baseline

  The correction write from the reinvestigation channel into the furnished file reads 2, and it is the structural claim of this board. Federal law makes that reinvestigation compulsory rather than discretionary, gives it thirty days extendable to forty-five, and lodges it at the single upstream store — so one dispute that succeeds fixes the record for every platform sharing the vendor at once. That is a wider correction pathway than a screening board usually carries, and it is why this deployment is not a story about a check that does not exist. The counterweight is drawn on the same diagram: the pleaded adverse action ran the day after the report, and nothing in the statute compels a platform to revisit an access decision it has already made.

• baseline

  The reconciliation of an aggregated item against the court that issued it reads 1 rather than zero, and the reason is legal rather than operational. The compulsory reinvestigation reaches the source that generated a disputed item, so the pathway is reachable by right. It reads no higher because that read happens one disputed item at a time on the subject's initiative, and because the pipeline otherwise reads aggregations by design, which is the gap the January 2024 advisory opinion addressed before it was withdrawn on 12 May 2025.

• baseline

  The user-certification precondition is drawn as a bounded automated screen on the furnishing pathway and reads 1. The statute bars a reporting agency from furnishing an employment-purpose report at all unless the user certifies compliance with the notice and pre-adverse-action duties, so the gate is in force and binds every such furnishing. It reads no higher because it is a written representation about the user's own process, obtained once for a relationship rather than read against each report — bounded by construction, which is what this edge kind means. The 2021 class complaint pleads that this agency furnished reports to one platform without obtaining that certification. That allegation is unadjudicated, it is the only count in that case against the agency, and no value on this diagram is set from it.

• baseline

  The independent second read of a match reads zero on what the record states rather than on what it omits. The vendor's own published account of the path is an instant criminal check returning in under a second with automated adjudication reducing manual workflows by up to eighty percent, and the regulator's stated standard for this deployment class is that matching a record to a person on name alone does not assure maximum possible accuracy. No source in this record describes a second, differently built read of a match anywhere in the pipeline, and no error rate for this pipeline has ever been published by anyone.

• baseline

  No enforcement node is drawn, and that absence is load-bearing rather than an omission. Every adverse action in this record is executed by a platform customer inside its own deactivation logic, in a different organisation one hop downstream, and the dossier's boundary note places those dynamics with the platform-side deployment class rather than here. Drawing an automatic downstream action system on this diagram would move a different deployment's machinery onto this one. The platform teams appear as an operator class instead, because that is where the record puts the decision.

• baseline

  No external-boundary node and no egress pathway is drawn. Furnishing a consumer report to a user with a permissible purpose is the product under a statutory authorisation regime, not data crossing a boundary without guardrails, and treating it as egress would count the deployment's ordinary operation as a leak. The one candidate crossing in the record — the precondition the statute puts on furnishing at all — is drawn as a bounded check on that pathway instead. No data-protection breach, regulator finding or adjudicated violation appears anywhere in this record.

• assumed

  No worklist, retriever or guardrail node is drawn. Nothing in this record describes a queue, a backlog or a triage list; nothing retrieves on demand, because the pipeline reads a pre-assembled aggregation, which is exactly the property the regulator's stated concern turns on; and the bounded screen this deployment does have is the statutory certification precondition, which is drawn as its own edge on the furnishing pathway.

• baseline

  Demand reads 3 from counted volume rather than from a default. The vendor's own marketing puts 140,000-plus business customers on the platform, ninety percent of gig-economy background checks running through it, and 89 percent of national criminal checks completing within one hour. The litigation supplies an independent denominator in one city: approximately 80,000 licensed rideshare vehicles in New York City, a platform policy re-checking incumbent drivers at least every two years, and pleaded classes of at least 1,000 and at least 300 since January 2020.

• baseline

  Capacity reads 2 because the manual comparator here is real and unmeasured at the same time. The statutory reinvestigation IS the human process — a person reading a disputed item against the court record that generated it — it is compulsory, it is specified, and it is what actually corrects a file, which rules out the low rung. Against that, no source measures the accuracy of that human process, no head-to-head comparison against this pipeline exists anywhere, and the volume a manual path would face at this scale is one no manual process has ever served, which rules out the high rung.

• baseline

  Every scale, speed, share and automation figure on this diagram is a vendor marketing claim fetched on 28 August 2026 and is labelled as one wherever it is used: ninety percent of gig-economy background checks, 140,000-plus business customers, 260 million identities covering 96 percent of United States adults, an instant check in under a second, 89 percent of national criminal checks within one hour, and automated adjudication reducing manual workflows by up to eighty percent. They are self-serving in both directions — capability marketing and an admission of how unmanned the generation path is — and no independent audit of any of them exists in this record.

• baseline

  Everything vendor-specific in this record is an allegation, and the two pleadings carry two different mechanisms that this file does not merge. No court has reached the merits and no regulator has taken public action against this vendor. In the 2021 case the underlying record was accurately reported so far as the pleading shows, and the pleaded violation against the agency is the missing user certification; the mis-association theory, that criminal case records plainly unassociated with the consumer were included and that deficient procedures are maintained because reporting more information is more profitable, belongs to the 2026 case and is pending. The order compelling individual arbitration could not be read from the verifying environment, and its treatment of the claims against the agency is asserted nowhere on this diagram.

• baseline

  The regulator channel reads 1 on the check pathway and 2 on the write pathway, and the pair is the measured contraction this case documents. Between November 2021 and January 2024 that channel wrote the operating rules for exactly this deployment class; on 12 May 2025 all four Fair Credit Reporting Act advisory opinions were withdrawn together inside a mass withdrawal of guidance. The statutes they interpreted are untouched and remain privately enforceable, so this file never says an accuracy duty was repealed or ended: what was removed is the regulator's stated reading of it. The check pathway carries the second half of the measurement, which is that no public enforcement action against this vendor was located at all.

• assumed

  Where this diagram diverges from the source model this network is derived from, it diverges in three places and asserts nothing that source does not already record. That source carries one source-pool store where this board draws an external feed plus the aggregation the pipeline reads, because the gap between them is the regulator's own concern. It has no edge kind for a check, so four of this board's checks redraw peer or read pathways from it, with their widths still taken from it. And it carries the courts as a governance actor rather than as a user class, so the private enforcement channel is drawn here from the litigation record directly. A fourth divergence this file recorded before is gone: the customer's eligibility matrix is still drawn as its own pathway, and the standing-monitoring product is now carried the way that source carries it, as a property of the file that is furnished and re-furnished.

• baseline

  Differential harm by race is recorded as an external observation and is never computed from anything on this diagram. The 2021 complaint pleads, citing a university centre's research, that 87 percent of New York City transportation independent contractors are persons of colour, 81 percent lack a college degree and 90 percent are foreign-born — a composition figure about the population this pipeline's errors land on, pleaded rather than adjudicated. The regulator stated separately that mistaken-identity risk from name-only matching is likely greater among Hispanic, Black and Asian communities because there is less surname diversity in those populations, which is a direction stated about a class of procedures rather than a measurement of this deployment. No per-cohort error rate for this deployment exists in any source.

What this example does not show

• Live and dominant vendor. Litigation: Golightly (S.D.N.Y.) compelled to individual arbitration Dec 21, 2022 and stayed, no public resolution located as of 2026-08-28; Davis (S.D. Fla.) pending, pre-certification; platform-side Aguilera v. Uber (Kings County) settled for $3.35M with Feb 2024 final-approval hearing. Regulator posture inverted in 2025: the CFPB advisory opinions that squarely covered this deployment were withdrawn May 12, 2025, leaving the FCRA statute itself as the operative constraint. No public CFPB or FTC enforcement action against Checkr itself was located.
• Screened workers are not modeled here. No report outcome, eligibility determination, deactivation or employment consequence for any person is computed from anything on this diagram; what propagates is institutional error through the operator network of the agency's matching and quality function, the platform teams that consume its reports, the statutory correction channel and the regulators. The person a report describes is the subject of it, and enters this network as the signal that opens a dispute.
• Every failure claim specific to this vendor is an allegation and stays in that register. No court has reached the merits and no regulator has taken public action against it. The 2021 complaint's only count against the agency is that it furnished employment-purpose reports without obtaining the user's compliance certification; the 2026 complaint pleads that criminal case records plainly unassociated with the consumer were included and that deficient procedures are maintained because reporting more information is more profitable. Those are two different mechanisms in two different cases, and this board does not merge them.
• Every scale, speed, share and automation figure is the vendor's own marketing, fetched on 28 August 2026 and independently audited by nobody: ninety percent of gig-economy background checks, 140,000-plus business customers, 260 million identities covering 96 percent of United States adults, an instant check in under a second, 89 percent of national criminal checks within one hour, and automated adjudication reducing manual workflows by up to eighty percent. They are self-serving in both directions — capability marketing and an admission of how unmanned the generation path is.
• The 12 May 2025 withdrawal removed interpretive guidance, not law. The maximum-possible-accuracy duty, the reinvestigation duty, the public-record currency-or-notice duty and the bar on furnishing an employment-purpose report without the user's certification all remain in force and privately enforceable. This board never says an accuracy duty was repealed or ended; what was withdrawn is the regulator's stated reading of the statutes, and courts may still reach the same results.
• The regulator actions in this record are adjacent-market posture, never enforcement against this vendor or against gig-employment screening. The $15 million joint action concerned a rental-screening arm and the $5.8 million action concerned two people-search reporting agencies, both in 2023. No equivalent public action against this vendor was located. The one completed money resolution here is platform-side — $3.35 million in a New York state court over Uber Eats courier background-check denials, where the agency was not a defendant and no admission of liability was located.
• The fan-out is topology-derived rather than incident-documented. That one report can produce work bans at several platforms at once follows from a shared upstream vendor and the vendor's own market-share claim, and no high-tier source documents a specific multi-platform simultaneous deactivation. It is narrated conditionally here and no named person is attached to it.
• Differential harm by race is a recorded external observation and is never computed from these dynamics. The composition figures — 87 percent of New York City transportation independent contractors are persons of colour, 81 percent lack a college degree, 90 percent are foreign-born — are pleaded in the 2021 complaint from a university centre's research, and the elevated mistaken-identity risk from name-only matching is a direction the regulator stated about a class of procedures. No per-cohort error rate for this deployment exists in any source.
• The jurisdiction is entirely United States and the correction-loop mechanics are specific to the Fair Credit Reporting Act. The thirty-day reinvestigation clock, the certification precondition and the reporting windows do not generalise outside it, and the order compelling individual arbitration could not be read from the verifying environment, so its treatment of the claims against the agency is asserted nowhere on this board.

Sources and evidence

What this example rests on, claim by claim. Every entry resolves to the same ledger the Evidence Registry publishes.

• Checkr, Inc. is a consumer reporting agency founded in 2014 whose automated background-check platform sits one company upstream of the gig platforms that buy from it: it retrieves criminal items from county, state, and national sources, matches them to a named applicant, normalises the charge data, applies each platform customer's own eligibility matrix, and furnishes a report, after which the platform executes its own access decision. Every scale, speed, and automation figure for it is the vendor's own marketing, fetched from its product pages on 28 August 2026 and independently audited by nobody: 90 percent of gig economy background checks run on Checkr; 140,000-plus businesses already run on Checkr; 260 million-plus real identities mapped, covering 96 percent of US adults; instant criminal checks that screen users in under a second; 89 percent of national criminal record checks complete within one hour; Checkr AI normalises criminal charge data; automated adjudication tools reduce manual workflows by up to 80 percent; and continuous checks are offered as a standing-monitoring product. Those claims are self-serving in both directions — they are capability marketing and they are also the clearest available statement of how unmanned the generation path is. Lyft, Hyer, and GigSmart are named as gig clients on the vendor's own pages; the Uber relationship is documented litigation-side, where a defence-bar analysis records Uber moving its New York City driver checks to Checkr in mid-2017.

  empirical
  • Vendor Checkr, Inc. (2026). Background Check Services for Gig and Marketplace (product page, fetched 2026-08-28); and Checkr, Inc. homepage (140,000-plus business customers) https://checkr.com/use-cases/gig-marketplace
  • Trade press Hunton Andrews Kurth LLP (2021, April 14). Gig Employer Hit with Background Check Class Action (Employment and Labor Perspectives) https://www.hunton.com/hunton-employment-labor-perspectives/gig-employer-hit-with-background-check-class-action

• The federal consumer regulator wrote down what the Fair Credit Reporting Act requires of automated background screeners, and its statements are about the industry rather than about any one company. In an advisory opinion of 4 November 2021 (Fair Credit Reporting; Name-Only Matching Procedures, 86 FR 62468) the Consumer Financial Protection Bureau held that name-only matching does not assure maximum possible accuracy under 15 U.S.C. 1681e(b), stated that when background screening companies and their algorithms carelessly assign a false identity to applicants for jobs and housing they are breaking the law, warned that the risk of mistaken identities from name-only matching is likely to be greater among Hispanic, Black, and Asian communities because there is less surname diversity in those populations, and observed that because of the sheer scale of background screening activity even ostensibly low error rates can harm significant numbers of consumers. In two companion advisory opinions of 11 January 2024 (Background Screening, 89 FR 4171; File Disclosure, 89 FR 4167) it stated that screeners must prevent the reporting of expunged, sealed, or legally restricted records, ensure dispositions accompany any reported arrest or charge, prevent duplicative reporting, and honour per-item obsolescence windows such as the seven-year bar on non-conviction arrests, and that a consumer's file disclosure must include both the originating sources and any intermediary or vendor sources. An independent legal-bar reading of the January 2024 opinions records the same four accuracy-procedure requirements and the same file-disclosure source rule. No error rate, accuracy figure, or per-cohort disparity measurement for Checkr's pipeline has ever been published by anyone.

  empirical
  • Government Consumer Financial Protection Bureau (2021, November 4). CFPB Takes Action to Stop False Identification by Background Screeners (advisory opinion: Fair Credit Reporting; Name-Only Matching Procedures, 86 FR 62468) https://www.consumerfinance.gov/about-us/newsroom/cfpb-takes-action-to-stop-false-identification-by-background-screeners/
  • Government Consumer Financial Protection Bureau (2024, January 11). CFPB Addresses Inaccurate Background Check Reports and Sloppy Credit File Sharing Practices (advisory opinions: Background Screening, 89 FR 4171; File Disclosure, 89 FR 4167) https://www.consumerfinance.gov/about-us/newsroom/cfpb-addresses-inaccurate-background-check-reports-and-sloppy-credit-file-sharing-practices/
  • Trade press Ballard Spahr LLP, Consumer Finance Monitor (2024, January 16). CFPB issues two new FCRA advisory opinions on background screening reports and disclosure of credit files to consumers https://www.consumerfinancemonitor.com/2024/01/16/cfpb-issues-two-new-fcra-advisory-opinions-on-background-screening-reports-and-disclosure-of-credit-files-to-consumers/

• The correction channel that reaches this deployment is statutory rather than discretionary, and the record documents it being outrun. Under 15 U.S.C. 1681i a consumer reporting agency must reinvestigate a disputed item and correct the file, within 30 days and 45 with an extension; 15 U.S.C. 1681k imposes a currency-or-contemporaneous-notice duty on public-record items reported for employment purposes; and 15 U.S.C. 1681b(b)(1) bars furnishing an employment-purpose report at all unless the user certifies compliance with the notice and pre-adverse-action duties. Because the report is assembled and held at one upstream agency, a dispute that succeeds corrects the record for every platform that reads it. Against that clock, the 2021 class complaint pleads that Checkr furnished Uber the report and that Uber deactivated Job Golightly one day later without any notice, process, or communication. The same complaint records the recurrence conditions: approximately 80,000 TLC-licensed rideshare vehicles in New York City, the majority on Uber, and an Uber policy of re-checking current drivers at least every two years through Checkr. The vendor separately markets continuous checks, a standing-monitoring product that re-furnishes on new record events. No dispute volume, reinvestigation resolution rate, or reversal rate for this vendor appears in any public source.

  empirical
  • Government Class Action Complaint, Golightly v. Uber Technologies, Inc. and Checkr, Inc., No. 1:21-cv-03005 (S.D.N.Y., filed 8 April 2021), hosted by Mobilization for Justice https://mobilizationforjustice.org/wp-content/uploads/Golightly-Uber-and-Checkr-Filed-Complaint.pdf
  • Government Consumer Financial Protection Bureau (2021, November 4). CFPB Takes Action to Stop False Identification by Background Screeners (advisory opinion: Fair Credit Reporting; Name-Only Matching Procedures, 86 FR 62468) https://www.consumerfinance.gov/about-us/newsroom/cfpb-takes-action-to-stop-false-identification-by-background-screeners/
  • Vendor Checkr, Inc. (2026). Background Check Services for Gig and Marketplace (product page, fetched 2026-08-28); and Checkr, Inc. homepage (140,000-plus business customers) https://checkr.com/use-cases/gig-marketplace

• On 12 May 2025 the Consumer Financial Protection Bureau withdrew all four of the Fair Credit Reporting Act advisory opinions covering this deployment class — Name-Only Matching (86 FR 62468), Permissible Purposes (87 FR 41243), Background Screening (89 FR 4171) and File Disclosure (89 FR 4167) — inside a mass withdrawal of 67 guidance documents, confirmed on the Bureau's own withdrawn-guidance list. The statutes those opinions interpreted were untouched: 15 U.S.C. 1681e(b), 1681i, 1681k, and 1681b(b) all remain in force and privately enforceable, and courts may still reach the same results. What was withdrawn is the regulator's stated reading of them, not the duties themselves, and no source in this record describes the accuracy duty as repealed or ended. The withdrawal is a documented contraction of the interpretive layer over a deployment class that continued operating unchanged.

  empirical
  • Government Consumer Financial Protection Bureau (2025, May 12). Withdrawn Guidance (list of guidance documents withdrawn 12 May 2025, including Name-Only Matching 86 FR 62468, Permissible Purposes 87 FR 41243, Background Screening 89 FR 4171 and File Disclosure 89 FR 4167) https://www.consumerfinance.gov/compliance/guidance/withdrawn-guidance/
  • Government Consumer Financial Protection Bureau (2024, January 11). CFPB Addresses Inaccurate Background Check Reports and Sloppy Credit File Sharing Practices (advisory opinions: Background Screening, 89 FR 4171; File Disclosure, 89 FR 4167) https://www.consumerfinance.gov/about-us/newsroom/cfpb-addresses-inaccurate-background-check-reports-and-sloppy-credit-file-sharing-practices/

• Two class vehicles have reached this pipeline and neither has been adjudicated on the merits. Golightly v. Uber Technologies, Inc. and Checkr, Inc., No. 1:21-cv-03005 (S.D.N.Y., filed 8 April 2021), pleads five counts of which four are against Uber — the New York City Fair Chance Act, disparate racial impact, and the federal and New York consumer reporting statutes — and one is against Checkr: furnishing employment-purpose reports without obtaining the user's compliance certification under 15 U.S.C. 1681b(b)(1). It alleges that Job Golightly, a Black Bronx driver who had driven for Uber since 2014 averaging about $1,500 a week, was deactivated on or around 28 August 2020 one day after Checkr furnished his report, over a single 2013 Virginia speeding offence at 22 miles per hour over the limit — a misdemeanour under Virginia law and a civil infraction under New York law — with no pre-adverse-action notice, no copy of the report, no Article 23-A individualised analysis, and no three-business-day window, and that he learned the reason months later. No misreporting by Checkr is pleaded in that case; the record was accurate as pleaded and the failures alleged are process failures. The complaint estimates classes of at least 1,000 and at least 300 since 11 January 2020, and records, citing research from The New School's Center for New York City Affairs, that 87 percent of New York City transportation independent contractors are persons of colour, 81 percent lack a college degree, and 90 percent are foreign-born. On 21 December 2022 Judge Lewis J. Liman granted the motion to compel individual arbitration and stayed the claims; the order itself could not be read from the environment that verified this record, its specific treatment of the claims against Checkr is not asserted, and no public resolution has been located as of August 2026. Davis v. Checkr, Inc., No. 0:26-cv-60088 (S.D. Fla., filed 26 January 2026), pleads a different mechanism: a putative nationwide class under 15 U.S.C. 1681e(b) alleging that a Checkr report included criminal case records plainly unassociated with the consumer and that the company knowingly and willfully maintains deficient procedures because reporting more information is more profitable, covering consumers who within two years received a Checkr report that incorrectly included criminal case records belonging to someone else. It is pending and pre-certification. All of the above are allegations.

  empirical
  • Government Class Action Complaint, Golightly v. Uber Technologies, Inc. and Checkr, Inc., No. 1:21-cv-03005 (S.D.N.Y., filed 8 April 2021), hosted by Mobilization for Justice https://mobilizationforjustice.org/wp-content/uploads/Golightly-Uber-and-Checkr-Filed-Complaint.pdf
  • Advocacy Mobilization for Justice (2021, April 8). Uber Named in First Class Action Lawsuit Challenging Discrimination Against Drivers with Criminal History and Violations of the Fair Chance Act and Fair Credit Laws https://mobilizationforjustice.org/news-and-press/uber-class-action-lawsuit/
  • Trade press Hunton Andrews Kurth LLP (2021, April 14). Gig Employer Hit with Background Check Class Action (Employment and Labor Perspectives) https://www.hunton.com/hunton-employment-labor-perspectives/gig-employer-hit-with-background-check-class-action
  • Government Opinion and Order, Golightly v. Uber Technologies, Inc. et al., No. 1:21-cv-03005, Doc. 51 (S.D.N.Y. 21 December 2022) (Liman, J.); host returned HTTP 403 on 2026-08-28 and the order was not read https://law.justia.com/cases/federal/district-courts/new-york/nysdce/1:2021cv03005/557807/51/
  • Trade press Top Class Actions (2026, January). Class action claims Checkr misreported criminal records in background checks (Davis v. Checkr, Inc., No. 0:26-cv-60088, S.D. Fla., filed 26 January 2026) https://topclassactions.com/lawsuit-settlements/lawsuit-news/class-action-claims-checkr-misreported-criminal-records-in-background-checks/

• Federal enforcement in neighbouring screening markets sets the posture the Fair Credit Reporting Act establishes for this class of automation, and none of it is an action against Checkr. On 12 October 2023 the Federal Trade Commission and the Consumer Financial Protection Bureau settled with TransUnion's rental-screening arm for $15 million — $11 million in redress and a $4 million penalty — over duplicate eviction entries, misreported outcomes, sealed records that were not removed, and undisclosed third-party sources. On 11 September 2023 the Commission took $5.8 million from TruthFinder and Instant Checkmate over background reports marketed as most accurate that were assembled from sources disclaiming accuracy, with a Flag as Inaccurate button that triggered no investigation; the Commission's position there was that report assemblers marketing for employment and tenant screening are consumer reporting agencies bound by the maximum-possible-accuracy and permissible-purpose duties. No public Commission or Bureau enforcement action against Checkr itself was located as of 28 August 2026. The one completed money resolution in this record is platform-side: Aguilera v. Uber Technologies, Inc. d/b/a Uber Eats, No. 509275/2023 (N.Y. Sup. Ct., Kings County), settled for $3.35 million over allegations that Uber Eats used a flawed criminal background check process that unfairly denied individuals access to the platform and failed to provide the legally required disclosures, covering New York City couriers denied between 24 October 2015 and 28 July 2021, split into serious-conviction and less-job-related groups, with automatic payments and a final approval hearing on 8 February 2024. Checkr was not a defendant there and no admission of liability was located.

  empirical
  • Government Federal Trade Commission and Consumer Financial Protection Bureau (2023, October 12). Settlement to Require Trans Union to Pay $15 Million over Charges It Failed to Ensure Accuracy of Tenant Screening Reports (a DIFFERENT product; not about sanctions name screening) https://www.ftc.gov/news-events/news/press-releases/2023/10/ftc-cfpb-settlement-require-trans-union-pay-15-million-over-charges-it-failed-ensure-accuracy-tenant
  • Government Federal Trade Commission (2023, September 11). FTC Says TruthFinder, Instant Checkmate Deceived Users About Background Report Accuracy, Violated FCRA While Marketing Reports for Employee and Tenant Screening https://www.ftc.gov/news-events/news/press-releases/2023/09/ftc-says-truthfinder-instant-checkmate-deceived-users-about-background-report-accuracy-violated-fcra
  • Trade press Top Class Actions (2023). Uber Eats New York background check $3.35M class action settlement (Aguilera v. Uber Technologies Inc. d/b/a Uber Eats, No. 509275/2023, N.Y. Sup. Ct. Kings County) https://topclassactions.com/lawsuit-settlements/closed-settlements/uber-eats-new-york-background-check-3-35m-class-action-settlement/

• A single automated rule set applied uniformly and without human review produced tens of thousands of correlated wrongful fraud determinations in the documented Michigan MiDAS case — one flaw repeating at caseload scale rather than averaging out.

  empirical
  • Government Michigan AG, settlement of civil-rights class action (Bauserman, 2022) https://www.michigan.gov/ag/news/press-releases/2022/10/20/som-settlement-of-civil-rights-class-action-alleging-false-accusations-of-unemployment-fraud
  • Investigative IEEE Spectrum, Michigan's MiDAS unemployment system: Algorithm alchemy that created lead, not gold https://spectrum.ieee.org/michigans-midas-unemployment-system-algorithm-alchemy-that-created-lead-not-gold

Where this connects

Institutional pressures in this domain

• Workload surge — Demand outruns staffing; per-case attention shrinks and review becomes triage.
• Vendor opacity — The deploying institution cannot inspect the model, data, or update pipeline it is accountable for.
• Compliance over substance — Paper controls (sign-offs, checklists) satisfy audits while the behavior they describe erodes.
• Data & policy drift — The world, the intake process, and the rules change under a system trained on how things used to be — two mechanisms with different remedies: the statistical properties of what the system processes move (concept drift), or the mixture of inputs arriving in deployment differs from the mixture it was trained on (covariate shift).
• Reviewer bottleneck — One fixed-capacity checking stage sits between AI output and consequence; everything queues behind it.

All of them in context on the Hiring & employment screening AI domain page.

Levers available here and the patterns behind them

• Upgrade model — Improve the model
• Store less data — Data minimization
• Check copied records — Reconcile copied records
• Mark AI-written records — Provenance labeling
• Check with a second model — Cross-model verification
• Vet connections — Connection authorization
• Gate record entries — Human-in-the-loop write gating
• Understand the system — Understand the system
• Review on schedule — Oversight cadence & retrospectives
• Pause AI on alarms — Deployment circuit-breaker

Documented case histories

• Checkr gig-economy background screening
• A resume screener that learned the past's bias
• Vendor screening across thousands of employers (litigation live)
• Graduate-hiring AI with its audits on the record
• HireVue video assessment (vendor layer)
• The 1959 statute and the integrity video screen (Baker v. CVS Health)
• An internal promotion, a recorded screen, and a captioning request (D.K. charges against Intuit and HireVue)
• Aon pre-hire assessment suite (vendor's own tables)
• The cooperative audit: a paid source-code examination, and what happened to its verdict
• McHire and the 64-million-record custody exposure
• SiriusXM's iCIMS applicant screening
• The rule with no number to disclose
• The account goes dark at nine; the reason arrives on day twenty-six
• iTutorGroup Tutor Application Screen
• Meta Job-Ad Delivery: the guardrail and the layer below