Aon's three-instrument pre-hire assessment suite

What this models

This example runs on the Three-instrument-class pre-hire assessment suite network: 10 components and 24 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: 2 assumed · 10 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

• baseline

  D48-derived new org (Phase 6, hiring-employment-screening), RE-DERIVED 2026-09-22 at a coarser granularity under the legibility programme. REGISTER THIS FIRST, because it governs every value on this board: nothing in this record has been found by anyone. There is no regulator finding, no court finding, no settlement, no consent order and no independent audit. Every characterization of these assessments as discriminatory or as deceptively marketed is an allegation by a civil-liberties complainant in two federal filings, and the vendor disputes it, stating that its assessment solutions follow industry best practices as well as Commission, legal and professional guidelines and are intended to be fair to everyone. The figures that order this board are the VENDOR'S OWN, published in its own technical documentation and quoted at table and page level in the complaint: the number is the vendor's and the reading is the complainant's, and both attributions travel together wherever either is used.

• baseline

  TOPOLOGY, and what the 2026-09-22 re-derivation changed. Ten nodes, all documented, none decorative. ONE model node for the instrument suite: the record documents three separately built products with three different mechanics, but they are sold as one suite, they are wired to the same counterparties in the same directions, and the sources show none of them acting differently on the error. Their differences are differences of content and of evidence weight, and they are carried in the node's own description and in the two assumptions below rather than drawn as three sub-graphs. TWO operator classes because the sources document exactly two parties acting on the output: the party that authors the items, the constructs, the scoring, the manuals and the published claims, and the party that buys the instrument and sets the cut-off. THREE stores: the vendor's own technical manuals; everything the vendor publishes about what the assessments do, to buyers on its sales pages and to candidates in its test-taker guide; and the per-candidate score output. The wiring between the first two IS the case, which is why those two are never folded together. The sales pages and the test-taker guide ARE folded together, because the record documents the same fairness framing on both surfaces, because the previous drawing carried that as a propagation pathway between them, and because no instrument this deployment could pull reaches one surface and not the other — both audiences it reaches stay drawn. ONE enforcement node because the marketed function of these instruments is bulk removal before more resource-intensive hurdles and because the population that step removes receives no human review. THREE reviewers because the record documents three separately constituted external channels, two of which are documented DOING different amounts: one served charge produced a dated act by the vendor, and the consumer filing has produced nothing public at all.

• baseline

  THE FORUM-TO-PRODUCT MAPPING, stated here because getting it wrong is the defect this record produces most easily, and because this board narrates it rather than drawing it. The class-wide discrimination charges filed in late 2023 cover the adaptive personality instrument and the gamified working-memory instrument. The consumer-protection complaint filed on 30 May 2024 covers all three products including the video scorer. No pathway on this board runs from a product to a forum and none ever did: what the record documents is that each FILING placed particular documents before a particular authority, and that is where the mapping now lives — in the two filing pathways, in each forum's own description, and here. The manual, whose tables anchor the discrimination theory, is what the charges placed before the employment-discrimination forum; the published claims, whose sentences are the deception theory, are what the 30 May 2024 filing placed before the consumer-protection forum.

• baseline

  THE VIDEO PRODUCT CARRIES THE THINNEST PROVENANCE IN THE RECORD, and that is an evidence-weight difference rather than a difference in how it is wired. Its two-stage pipeline is real and documented, but the identification of which commercial speech-to-text service and which classifier family comes from a developer's published dissertation and a patent rather than from any vendor disclosure, and the speech-recognition error research the complaint leans on tested a system from the same provider rather than necessarily the deployed version, which the complaint itself flags. The earlier drawing expressed that weight by running the product's own pathways one rung below their siblings; this board expresses it in the copy on those pathways and here, because the pathways are otherwise identical. No node and no pathway on this board is drawn on that identification, and no provider is named anywhere in this bundle.

• baseline

  ABSENCES ARE DERIVED TOO, and four of them are load-bearing. There is NO input-source node for the machine transcript, on the provenance stated above: drawing a node on the thinnest identification in the record would put more weight on it than it carries, so the two-stage pipeline is stated in the suite's own description and in the pathway that runs the suite against itself. There is NO external-boundary node and no egress pathway, because no data-protection breach, disclosure finding or regulator determination exists anywhere in this record. There is NO contest channel for a screened-out person: the record documents none, and assessed people are boundary-only on every Lab diagram, so the absence is drawn as the reconciliation pathway that runs at zero. There is no worklist, no retriever and no guardrail: no queue, backlog, retrieval component or automated screen over the instruments' output appears anywhere in these sources.

• baseline

  WHERE THE LAB SHAPE DIVERGES FROM THE PAN SHAPE, and nothing is asserted here that the PAN file does not already record. Four divergences. First, the PAN entry carries ONE federal-forum user class; this board draws TWO, because the record shows them doing different amounts — one served charge is followed by a dated act and the consumer filing is followed by nothing public — and collapsing them would have required a single figure where the record carries two. The single peer pathway from the complaint channel into that class is drawn onto both nodes at its own width. Second, the PAN entry's three record stores are the manual, the sales channel and the candidate guidance, and none of them is the per-candidate output; the Lab folds the last two into one published-claims store and adds a score record and an enforcement node, because the complaint describes the outputs and the marketed screen-out function directly, and because the PAN entry's own deployment factors record human review as false, human-in-the-loop as false, autonomous determination as true and the system as gating. Third, the PAN entry has no representation for a check at all, so two of this board's four checks are its pathways redrawn — the charge reaching the vendor, and the requested relief acting on the published claims — with their widths still taken from the source on the stated mapping. Fourth, the PAN entry carries no operator-into-model pathway, so the buyer's cut-off dial is Lab-side and is derived from the complaint's own concession at paragraph 62.

• baseline

  BASELINES, and what the re-derivation did and did not do to them. NO baseline was invented, raised or lowered to fit a smaller board. Where two pathways were folded into one, the survivor keeps the rung the wider of them already carried, and the narrower sibling's rung is stated in the survivor's own copy: the published guidance into the instruments keeps 2 with the video product's narrower counterpart noted; the manual's specification pathway keeps 3, which is the superlative the source asserts; the score pathway to the hiring team keeps 3 for the same reason; the vendor's write of its published claims keeps 3, the widest write in the record and the one nothing has corrected. Four superlatives the source states in its own words survive the translation unchanged: the widest read (the manual read by its authors), the widest write (the vendor writing its claims), the thinnest read (the manual read by the buyer), and the one pathway the source says is drawn at its floor deliberately (the manual into the published claims). Three contrasts carry the reading. The published claims reach the buyer at the top rung while the manuals that report the group differences reach the same buyer at the bottom rung — that pair is the unfairness theory drawn as two numbers. The pathway from the manuals into the published claims runs at zero while the reconciliation back the other way also runs at zero — nothing carried the evidence into the claim and nothing checked the claim against the evidence, which is the deception theory as a picture rather than as a complaint. And the score record drives the screen at the top rung while the reading of that screen back against the record runs at zero, which is what the phrase 'no human review' means when it is drawn instead of narrated. Five pathways run at zero and every one of them is a documented absence rather than an unexamined one; none was removed to make this board smaller.

• baseline

  DEMAND 3 / CAPACITY 1, unchanged by the re-derivation. Demand 3 on the vendor's own marketing figures, labelled as such: more than 30 million assessments a year, more than 40 languages, 90 countries, with a third-party market-data claim of the second-largest global share quoted in the complaint. Capacity 1 is the unusual half. The counterfactual floor is the human process that would run with no instrument at all, and this record describes that floor as exactly what the product is sold to displace: the assessments are marketed to screen out applicants prior to more resource-intensive hurdles such as resume screening and interviews, and one client case study the vendor publishes is cited for screening out 62 percent of applicants on the assessment alone. For the population removed at that step the dossier's structural statement is that the assessment is the first and the final decision point and that no person reads their file. A manual comparator that the seller's own pitch identifies as too expensive to run at this volume, and that never sees the removed population at all, is a floor of 1. What the record does not contain is any measurement of how a human process would have performed on the same applicants; nothing on this board computes that, and the value rests on the documented marketed position of the instrument rather than on any measured comparison.

• baseline

  EVIDENCE STATUS, labelled where it is used, because this record mixes tiers unusually. VENDOR-PUBLISHED AND VENDOR-MEASURED: every figure at the centre of this board — the two-week test-retest coefficients with ten of fifteen constructs below the 0.7 floor and a minimum of 0.44, the group score differences of 0.48, 0.38, 0.37 and 0.35 standard deviations on the gamified instrument, and the sibling-instrument effects of 0.86 and 1.21 — comes from the vendor's own technical documentation, quoted in the complaint. VENDOR MARKETING: the scale figures, the screen-out positioning, and the 62 percent client case study. VENDOR CLAIM, LIVE-FETCHED: the pre-hire page language recorded on 28 August 2026. VENDOR STATEMENT: the response through trade press in June 2024. ADVOCACY FILINGS: the two federal filings and every characterization in them. LEGAL JOURNALISM: the status checks of March and August 2025 and the record of the enforcement-environment change. MANAGEMENT-SIDE LEGAL ANALYSIS: the procedural framing of what each filing does and the confirmation of the forum-to-product scope. No parameter on this board is scaled by a vendor figure without that figure being named as the vendor's on the line that uses it, and no allegation is scaled as though it were a finding.

• assumed

  AN AVERAGE SCORE DIFFERENCE IS NOT AN ADJUDICATED ADVERSE IMPACT, and the complaint concedes it. Real selection impact depends on the cut-off scores and the tool combinations the employer sets, which the vendor does not control. This board therefore draws the gap-to-impact translation as a governance surface held by the buyer — one operator-into-model pathway, at the only place a threshold can be moved — and asserts no violation anywhere. The disability theory runs through construct overlap rather than through a measured screen-out: statements the charging party recalled encountering closely parallel items on two clinical autism screeners, the vendor's own interpretation guides flag extreme-end scorers as potential problems for employers, meta-analyses cited in the complaint show autistic people scoring lower on working-memory measures of the kind the gamified instrument uses, and a 166-page manual documents no disability-bias measure beyond an outside attorney's review of item wording limited to physical-disability terms. What is alleged is a high RISK of screening out disabled applicants, plus one charging party's individual account. No adjudicated or statistically demonstrated screen-out of disabled applicants exists in this record and none is asserted here.

• baseline

  THE STORE'S DOCUMENTED BLIND SPOT, stated as the structural fact it is. The material that would show whether these instruments select well was never gathered: no error rate, accuracy figure or defect rate for any of the three products has ever been published by anyone, and no deployment of any of them has been independently evaluated. The sole systematic studies are the vendor's own, reported in the vendor's own manuals, and the sole documented disability-bias measure among them was an outside attorney's review of item wording. That is why the check that would evaluate these instruments from outside runs at zero rather than low, and why the absence is a finding of this case rather than a gap in the model. Nothing on this diagram computes a harm to a screened-out person, and nothing could.

• assumed

  Assessed people are not in the dynamics. No hiring decision, screen-out, score or employment outcome for any person is computed from anything drawn here, and no score over any person is authored anywhere in this network. The group score differences are recorded external observations: they are mean differences in standard deviations, reported by the vendor about its own study samples and quoted by a complainant, and they are neither selection-rate ratios nor error rates nor findings of discrimination. The manuals that carry them were removed from public access after the charges were filed and could not be independently re-verified; the complainant cites archived copies and states it knows of no other public source. The employer co-respondent on the discrimination charges is publicly unnamed and is named nowhere in this bundle, and the vendor's marquee customers are not that respondent.

What this example does not show

• LITIGATION AND REGULATORY POSTURE, verbatim from the evidence dossier and load-bearing. Both forums open-but-silent as of 2026-08-28: the FTC complaint is an advocacy request for investigation that the agency has never publicly acted on; the EEOC class charges have produced no publicly visible resolution, determination, or follow-on lawsuit (EEOC charge proceedings are confidential, so 'pending' is an inference from the absence of any public disposition). The federal enforcement environment shifted sharply against both theories during the pendency (guidance withdrawals Jan 2025; disparate-impact deprioritization EO Apr 2025). Aon continues to sell all three products and continues materially similar fairness marketing on live pages.
• EVERYTHING CHARACTERIZING THESE ASSESSMENTS AS DISCRIMINATORY OR AS DECEPTIVELY MARKETED IS AN ALLEGATION. No regulator, court or independent auditor has made any finding of any kind, and there is no lawsuit, no right-to-sue notice on this record, no settlement and no consent order. The vendor disputes the claims and its position is carried wherever they are summarized: its assessment solutions follow industry best practices as well as Commission, legal and professional guidelines and are intended to be fair to everyone. In August 2025 an Aon spokesperson declined further comment. The phrase 'deceptively marketed' is the complainant's claim and is never this catalogue's voice.
• THE ADVERSE-IMPACT NUMBERS ARE THE VENDOR'S OWN, quoted by a complainant. They come from Aon's technical documentation as reproduced in the ACLU's FTC complaint: the gridChallenge and GAME manual's Table 43 at 0.48, 0.38, 0.37 and 0.35 standard deviations, and sibling tools at 0.86 and 1.21. Both attributions travel together — the number to Aon, the reading to the ACLU. The underlying manuals were removed from public access after the EEOC charges were filed and could not be independently re-verified for this record; the complaint cites archived copies and the complainant states it knows of no other public source.
• AN AVERAGE SCORE DIFFERENCE IS NOT AN ADJUDICATED ADVERSE IMPACT, and the complaint concedes it at paragraph 62: real selection impact depends on the cut-off scores and the tool combinations the employer sets, which the vendor does not control. This board draws that translation as a governance surface held by the buyer and asserts no four-fifths-rule violation, no disparate impact and no discrimination anywhere. The complaint's own translation aid — that even small average differences can produce four-fifths-rule problems at commonly occurring selection ratios — is cited in the complaint as a general finding about selection statistics and is not a measurement of this deployment.
• THE DISABILITY THEORY RUNS THROUGH CONSTRUCT OVERLAP, NOT THROUGH A MEASURED SCREEN-OUT. What is alleged is a high RISK of screening out autistic and disabled applicants, plus one charging party's individual account. No adjudicated or statistically demonstrated screen-out of disabled applicants exists in this record and none is asserted. The ADEPT-15 statement wordings attributed to the test are the charging party's recollections and the precise wording may differ.
• THE FORUM-TO-PRODUCT MAPPING, stated because it is easy to get wrong and this board is built on it. The EEOC class charges cover ADEPT-15 and gridChallenge. vidAssess-AI appears in the FTC complaint and does not appear in the EEOC charges. Nothing in this scenario places the video product before the discrimination forum. The mapping is carried in the copy on the two filing pathways and in each forum's own description rather than drawn as structure: no pathway on this board runs from a product to a forum, and none ever did.
• THE VIDEO PRODUCT'S PIPELINE CARRIES THE THINNEST PROVENANCE IN THE RECORD. The speech-to-text provider and the classifier family come from a developer's published dissertation and a patent rather than from any Aon disclosure, and the speech-recognition error research the complaint cites tested a system from the same provider rather than necessarily the deployed version — which the complaint itself flags. No node on this board is drawn on that identification, no provider is named anywhere in this bundle, and the compounding argument is carried inside the suite's own description and on the pathway that runs the suite against itself, rather than as an asserted supply chain.
• NEITHER FORUM'S SILENCE MEANS ANYTHING IN EITHER DIRECTION. The consumer-protection authority never docketed a public matter, which means no PUBLIC action and cannot rule an unannounced staff inquiry in or out. Charge proceedings before the employment authority are confidential, so the charge could have been resolved, settled, withdrawn or dismissed without public trace; 'no public resolution located as of 28 August 2026' is the maximal honest claim, and 'pending' is never stated as a positive fact. The post-2025 enforcement retreat explains the silence without implying that the filings lacked merit — or that they had it.
• THE CLIENT EMPLOYER CO-RESPONDENT IS PUBLICLY UNNAMED, described only as a mid-sized company headquartered in the United States, and is named nowhere in this bundle. Aon's marquee customers listed on its own site and in third-party lists quoted in the complaint are NOT that respondent, and no employer is presented as one. No cut-off, vetting practice or configuration for any named client appears anywhere in this record.
• ASSESSED PEOPLE ARE NOT MODELED. No hiring decision, screen-out, score or employment outcome for any person is computed from anything on this diagram, and no score over any person is authored anywhere. The group score differences are recorded external observations of mean differences in standard deviations reported by the vendor about its own study samples; they are not selection-rate ratios, not error rates and not findings of discrimination. The structural fact the record does establish — that a screened-out candidate receives no human review — is drawn as a reconciliation pathway that runs at zero, never as a harm anyone has counted.

Sources and evidence

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

• Aon Consulting, Inc., the human-capital arm of Aon plc, designs, markets, and administers a suite of pre-hire assessments sold into other companies' hiring pipelines. Three of them are at issue: ADEPT-15, a computer-adaptive forced-choice personality test scoring fifteen constructs from item pairs matched for social desirability out of a bank of more than a thousand statements, with no skip option and no agree-with-both option, whose outputs are construct scores, job-fit scores, and generated interview scripts for the employer; gridChallenge, a gamified working-memory test of nine items each pairing a timed target-icon memorisation task with a symmetry or shape-judgment distractor task; and vidAssess-AI, an asynchronous video product that transcribes a candidate's spoken answers with a commercial speech-to-text service and scores phrases in that transcript as far-extreme positive or negative indicators of the same ADEPT-15 constructs. On Aon's own marketing, quoted in the ACLU's complaint, the company administers more than 30 million assessments a year in over 40 languages across 90 countries, and third-party market data quoted in the same paragraph places it at the second-largest share of the global pre-hire assessment market. The assessments are marketed to screen out applicants prior to more resource-intensive hurdles such as resume screening and interviews, and Aon's own case-study marketing is cited for one client that screened out 62 percent of applicants on the assessment alone; for the population removed at that step there is no human review, and the assessment is the first and the final decision point. Every scale and screen-out figure here is Aon's own marketing and none is independently audited.

  empirical
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Advocacy American Civil Liberties Union (2024, May 30). ACLU Files FTC Complaint Against Major Hiring Technology Vendor for Deceptively Marketing Online Hiring Tests as 'Bias Free' https://www.aclu.org/press-releases/aclu-files-ftc-complaint-against-major-hiring-technology-vendor-for-deceptively-marketing-online-hiring-tests-as-bias-free
  • Trade press HR Dive (2024, June 5). ACLU asks FTC to probe Aon AI employment assessment tools for bias https://www.hrdive.com/news/aclu-ftc-complaint-aon-ai-employment-assessment-tools-biased/717954/

• The figures at the centre of this record are Aon's own, published in Aon's technical documentation and quoted at table and page level in the ACLU's complaint to the Federal Trade Commission; the numbers are Aon's and the reading of them is the ACLU's. On ADEPT-15, the reliability table shows ten of fifteen constructs producing two-week test-retest coefficients below the 0.7 floor Aon's own documents tell buyers to demand, the lowest being 0.44 on 'Awareness', which the complaint identifies as the construct with the greatest overlap with autism diagnostic screeners. On gridChallenge, the gridChallenge and GAME test documentation at Table 43 reports Black assessment-takers scoring on average 0.48 standard deviations below white assessment-takers, with two-or-more-ethnicities at 0.38, Hispanic or Latino at 0.37 and Asian at 0.35; Aon's documentation for sibling cognitive tools reports Black test-takers averaging 0.86 standard deviations lower on scales clues and 1.21 lower on switchChallenge, which are 'large' effects under the Cohen's-d heuristic Aon itself applies in the same documents. An average score difference is NOT an adjudicated adverse impact: the complaint concedes at paragraph 62 that actual selection impact depends on the cut-off scores an employer sets and on how the assessment is combined with other tools, which Aon does not control. No error rate, accuracy figure, or independent evaluation exists for any of the three products, from any source. The manuals these figures come from were removed from public access after the EEOC charges were filed and could not be independently re-verified; the ACLU states it knows of no other public source and cites archived copies. Aon disputes the complaint and states that its assessment solutions follow industry best practices as well as the Equal Employment Opportunity Commission, legal, and professional guidelines and are intended to be fair to everyone.

  empirical
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Trade press HR Dive (2024, June 5). ACLU asks FTC to probe Aon AI employment assessment tools for bias https://www.hrdive.com/news/aclu-ftc-complaint-aon-ai-employment-assessment-tools-biased/717954/

• The ACLU's disability theory runs through construct overlap rather than through any measured screen-out. Statements the EEOC charging party recalled encountering on ADEPT-15 — among them 'I tend to avoid large groups of people' and 'I have difficulty determining how someone feels by looking at their face' — are set in the complaint's Table 1 beside near-parallel items on the clinical autism screeners AQ and RAADS-R; the statement wordings are the charging party's recollections and precise wording may differ. Aon's own interpretation guides flag extreme-end scorers as potential problems for employers. Meta-analyses cited in the complaint show autistic people scoring significantly lower on working-memory measures of the kind gridChallenge uses. And the design record is the load-bearing part: across a 166-page ADEPT-15 technical documentation, the only disability-bias measure the complaint identifies is an outside-attorney review of item wording that flagged physical-disability language such as 'seeing', 'speaking', and 'hearing' — with no clinical or neurodiversity sensitivity review and no study on disabled populations anywhere in the programme. What the complaint alleges is a high RISK of screening out autistic and disabled applicants, together with one charging party's individual account. No adjudicated or statistically demonstrated screen-out of disabled applicants exists in this record. In August 2025 Bloomberg Law reported the EEOC charge as filed and unresolved and recorded that an Aon spokesperson declined to comment.

  empirical
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Trade press Bloomberg Law, Daily Labor Report (2025, August 22). AI Hiring Tools Elevate Bias Danger for Autistic Job Applicants https://news.bloomberglaw.com/daily-labor-report/ai-hiring-tools-elevate-bias-danger-for-autistic-job-applicants

• Two federal forums were opened on one product suite and neither has publicly moved. In late 2023 the ACLU, with co-counsel Winston Cooks, LLC, filed class-wide charges of discrimination with the Equal Employment Opportunity Commission under the Americans with Disabilities Act and Title VII against Aon AND a client employer, on behalf of a biracial Black and white autistic job applicant with mental-health disabilities; the employer is publicly identified only as a mid-sized company headquartered in the United States. Those charges cover ADEPT-15 and gridChallenge. On 30 May 2024 the ACLU filed a complaint and request for investigation with the Federal Trade Commission against Aon Consulting, Inc., alleging that marketing the assessments as bias-free, fair, having no adverse impact, and able to improve diversity is a deceptive act or practice under Section 5 of the FTC Act, and that selling assessments alleged to discriminate while failing to assess disability harms is an unfair practice injuring both workers and employers; that complaint covers all three products including vidAssess-AI. The relief requested asks the Commission to open an investigation, enjoin the claims, require truthful risk information, and require Aon to pause sale or administration of the assessments until discrimination is eliminated or discontinue products where it cannot be. Management-side analysis of the filings records the procedural register: an advocacy complaint to the Commission initiates agency review and triggers no automatic enforcement, and an EEOC charge is confidential while pending and is commonly a precursor to private litigation. The enforcement environment contracted during the pendency: in January 2025 the EEOC removed its artificial-intelligence hiring technical-assistance guidance following the administration change, and an executive order of 23 April 2025 directed federal agencies to deprioritize disparate-impact liability enforcement, the theory underlying the race claims; legal-press analysis in March 2025 listed the Aon matters among pending ones with no indication of closure. As of 28 August 2026 no public FTC action, no public EEOC determination, settlement or follow-on lawsuit, and no product withdrawal has been located. Because charge proceedings are confidential and the Commission never docketed a public matter, that silence establishes nothing in either direction: it means no PUBLIC action, and an unannounced staff inquiry can be neither ruled in nor ruled out.

  empirical
  • Advocacy American Civil Liberties Union (2024, May 30). ACLU Files FTC Complaint Against Major Hiring Technology Vendor for Deceptively Marketing Online Hiring Tests as 'Bias Free' https://www.aclu.org/press-releases/aclu-files-ftc-complaint-against-major-hiring-technology-vendor-for-deceptively-marketing-online-hiring-tests-as-bias-free
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Trade press Fisher Phillips (2024). AI Hiring Tools Under Attack: ACLU Files Claims with Feds Over Common Hiring Tools https://www.fisherphillips.com/en/news-insights/ai-hiring-tools-under-attack-aclu-files-claims-with-feds-over-common-hiring-tools.html
  • Trade press Bloomberg Law, Daily Labor Report (2025, March 26). Trump's Disparate Impact Blow Makes AI Bias Claims Even Tougher https://news.bloomberglaw.com/daily-labor-report/trumps-disparate-impact-blow-makes-ai-bias-claims-even-tougher

• The contested link in this record is between two sets of documents the same company wrote. The ACLU's complaint quotes Aon's marketing as describing the assessments as 'fair[] for all: no adverse impact', 'scientifically proven not to have bias', 'culturally-agnostic', and as excluding prejudices, and alleges that selling on those sentences while Aon's own technical documentation reported the group score differences above is a deceptive act under Section 5. Aon responded through trade press in June 2024 that its assessment solutions 'follow industry best practices as well as the Equal Employment Opportunity Commission, legal and professional guidelines' and are 'intended to be fair to everyone'; that is a vendor statement and no adjudication of any of it exists. The record contains one documented act by Aon after the EEOC charges were filed, and it went to evidence rather than to claims: the public link to the assessment technical documentation on Aon's Norwegian support portal was disabled, and the ACLU states it knows of no other public source and cites archived copies. The marketing did not change. When Aon's pre-hire talent assessment page was fetched on 28 August 2026 it advertised 'Fair and unbiased assessments' and asserted that 'Assessment tools that are scientifically proven not to have bias are essential' — twenty-seven months after those claims were formally alleged to be deceptive, with all three products still on sale. That page is recorded as a vendor claim, its content can change, and the fetch date is the citation of record.

  empirical
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Vendor Aon plc (2026). Pre-Hire Talent Assessment (live product marketing page, fetched 28 August 2026) https://www.aon.com/en/capabilities/talent-and-rewards/pre-hire-talent-assessment
  • Trade press HR Dive (2024, June 5). ACLU asks FTC to probe Aon AI employment assessment tools for bias https://www.hrdive.com/news/aclu-ftc-complaint-aon-ai-employment-assessment-tools-biased/717954/

• Two documented properties of this deployment decide what an assessed person can do about a score. First, the position of the instrument: the assessments are marketed to screen out applicants prior to more resource-intensive hurdles, and for the population screened out there is no human review at all — the assessment is the first and the final decision point, and Aon's own case-study marketing is cited for one client that screened out 62 percent of applicants on the assessment alone. Second, what the candidate is told before responding: Aon's test-taker guide states that 'there is no possibility of bias — for or against specific candidates — either during the test or during the evaluation of answers', and discusses accommodations only for physical disabilities. The ACLU argues that a candidate who reads that does not make the accommodation request that would generate the only disability-relevant signal anywhere in the system, and that the absence of the signal then reads as evidence that nothing was needed; invoking the request costs the candidate a disclosure of disability. The quoted text is Aon's and the alleged suppressing effect is the ACLU's. The employer, not Aon, holds the decision that converts a score into an outcome: which constructs are used, where the cut-off falls, and what the assessment is combined with. No cut-off or vetting practice for any named client appears in this record, and the client employer named as an EEOC co-respondent is publicly unnamed.

  empirical
  • Advocacy American Civil Liberties Union Foundation (2024, May 30). In the Matter of Aon Consulting, Inc.: Complaint and Request for Investigation, Injunction, and Other Relief, filed with the Federal Trade Commission under Section 5 of the FTC Act https://www.aclu.org/wp-content/uploads/2024/05/In-re-Aon-Consulting_FTC-Act-complaint_052924.pdf
  • Advocacy American Civil Liberties Union (2024, May 30). ACLU Files FTC Complaint Against Major Hiring Technology Vendor for Deceptively Marketing Online Hiring Tests as 'Bias Free' https://www.aclu.org/press-releases/aclu-files-ftc-complaint-against-major-hiring-technology-vendor-for-deceptively-marketing-online-hiring-tests-as-bias-free

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

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

Documented case histories

• Aon pre-hire assessment suite (vendor's own tables)
• 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)
• 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
• Checkr gig-economy background 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