PAN Lab example

Intuit's recorded video assessment for promotion

Assessing an incumbent: a recorded promotion gate and the captioning request

A large technology employer runs an internal promotion through a standardised recorded assessment. A seasonal employee of five years applies for the next role up. She has already been promoted once, she leads a team supporting about four hundred associates, she has earned a bonus every year, her supervisors have praised her communication, and the manager who encouraged her to apply sits on the hiring team. The employer, in other words, already holds five seasons of direct measurement of the person the assessment is there to estimate. She is invited into the assessment step anyway: about a dozen timed recorded video questions, plus essay and multiple-choice sections, roughly three hours in total, with instructions and questions delivered audibly. She is Deaf. As alleged in a Complaint of Discrimination filed on 19 March 2025 with the Colorado Civil Rights Division and the EEOC, she asked for human-generated real-time captioning — deliberately not sign-language interpretation, because interpreters had not handled tax concepts well before. She was told the platform's built-in subtitles could be turned on. When the assessment began, she says, there was no subtitle option. She completed three hours of it on browser automatic captions she supplied herself, which she says mis-transcribed complicated words. Being offered a version without the recorded video interview was never on the table, and she says she feared asking. That sequence is the whole of the employer-side allegation, and it does not depend on what happened to her answers afterwards: a process existed, a request went through it, and the answer named a control that was not there. What happened to the answers is the part the parties dispute at its foundation. The charge pleads a three-stage pipeline — automatic speech recognition of the spoken answers, a second stage interpreting the transcript, machine-learning scoring against job competencies — and pleads it on information and belief, inferring it from a single message. The vendor's chief executive answers that the complaint is entirely without merit, that it rests on an inaccurate assumption about the technology used, and that the employer did not use one of the vendor's artificial-intelligence-based assessments. The employer separately says the allegations are entirely without merit and that it provides reasonable accommodations to all candidates. Nothing has been found. Under the vendor's version, the same facts describe human reviewers who let a three-hour screen's result stand over five seasons of their own evidence. Two things reached her. An automated rejection on 13 August 2024, saying the employer had decided to move forward with other candidates. Then, on 25 September, a feedback message advising more concise and direct answers, adapting her communication style to different audiences, and practising active listening. She believed from its return address and register that a machine wrote it. On 31 October she exercised her state-law right to inspect her personnel file, to find out how the decision had been made; the charge alleges only minimal documents were produced. The employer had been told. After the previous season she had joined its own accessibility advisory body and warned the body's chair that this assessment step could exclude deaf applicants. She was told it would be looked into. Nothing was reported back, and the following year the same step was required of her. Four years before that, per the charge, the employer's own automated call-monitoring metric had scored how closely agents followed a script from speech-recognition transcripts and read her deaf accent as departure from it, producing one artificially low indicator against otherwise strong measures. The employer moved her from phone work to chat. The indicator stayed in her file. The published research the charge rests on measured four widely used general-purpose speech-recognition services and found a mean word error rate of 52.6 per cent for deaf and hard-of-hearing speakers against 5.0 per cent for normal-hearing ones — a measurement of those four services, not of this platform and not of this interview. Before you pick a target level: this board cannot be won under Service and Safety Targets or All Governance Targets, and you cannot buy your way past them. Take every instrument this employer could actually reach, set each one to full strength and ignore the budget entirely, at a total of fifty against the six you are given. Four pathways are still open at the end. They are the recorded session being the thing the assessment reads, the employer deciding that a promotion will run through that assessment at all, an employee's own record becoming the file the law lets her ask for, and the answer reaching the person it is about. Those four are not a gap in this deployment's governance. They are what an assessment-gated promotion is: a session, a decision to use it, a record, and a message. Closing all four would not be a better-governed version of this deployment. It would be its absence. That is a measurement of the deployment this network is drawn from, rather than a puzzle waiting to be cracked. Explore and Service Targets Only can both be won, and cheaply: two instruments, costing five of your six. Note as you play that piling on more is not the same as doing better here, because the widest stacks cost this deployment the benefit it produces.

Stylized model of a documented deploymentHiring & employment screening AI

Open this example in PAN Lab v0.1 to apply pressures and levers and watch what the system does.

What this models

This example runs on the Internal-promotion-class recorded assessment gate network: 11 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: 6 assumed · 8 published baseline. In the Lab, the shaded evidence band behind each headline readout draws its width from the least-established class below.

baseline
This deployment is a pending administrative charge and nothing in it has been found. Every operative statement drawn here about what happened to the complainant is an allegation of a Complaint of Discrimination filed 19 March 2025 with the Colorado Civil Rights Division and the EEOC, or the complainant's own published account of April 2025. Both respondents dispute the charges. No probable-cause determination, dismissal, right-to-sue notice, court filing or settlement had surfaced by 28 August 2026, and administrative investigations are confidential, so that silence establishes nothing in either direction.
baseline
The vendor disputes the mechanism itself, not only the legal conclusions, and this board carries that at the level of the pathway rather than as a footnote. Its chief executive states that the complaint rests on an inaccurate assumption about the technology used and that the employer did not use one of the vendor's artificial-intelligence-based assessments. Under that account the assessment component composed nothing and the recorded session was reviewed by people — which changes what travels along the pathway into the panel without changing that the decision ran through it.
baseline
The two pleaded layers have different evidentiary footings and the diagram keeps them apart. The accommodation-denial pathway is alleged against the employer and survives either resolution of the technical dispute; the scoring pathway depends on it. That is why the accommodation disposition reaching the session is drawn as its own width, and why the assessment component's description carries the denial in the same sentence as the allegation.
baseline
The speech-recognition measurement stays attached to the study that made it. A peer-reviewed study measured four widely used general-purpose commercial speech-recognition services on a twenty-four-speaker corpus and reported a mean word error rate of 52.6 per cent for deaf and hard-of-hearing speakers against 5.0 per cent for normal-hearing speakers, rising to 85.9 per cent for the lowest speech-intelligibility group. It did not measure this platform, this interview or any component drawn here. No width on this board is a rate for this deployment, and no error rate for either component exists from any source.
baseline
The employer's call-monitoring metric is drawn as a separate component rather than folded into the promotion screen, because the record keeps them separate and so must the diagram. It belongs to the employer and not to the interview platform, it ran four years earlier, and the charge uses it as evidence of what the employer already knew rather than as the assessment that denied the promotion. The pathway between them is the check drawn at zero, which is the notice theory rather than a data flow.
assumed
The recorded session is drawn as its own input rather than folded into the assessment record, which is where the PAN entry for this deployment keeps it. The split is disclosed because it is a Lab-side refinement: in this deployment the session is also the access surface the accommodation request was about, so folding it would hide the link where the employer-side allegation lives.
assumed
The automated notices are drawn as a downstream action system, which the PAN entry does not carry as a separate component. The reason is that they are the only part of this deployment that ever reached the assessed person: an automated rejection and a feedback message. Everything anyone outside knows about this screen was reconstructed from one of them. The automated character of the rejection is pleaded against the employer's process; whether the feedback message was machine-generated by the vendor's analysis is the inference the vendor disputes, and both are carried without either being asserted as found.
baseline
Four widths are drawn at zero and each rests on a documented absence rather than on silence. The accommodation disposition reached no control in the session, because the subtitle option it named was, as alleged, not present when the session began. Nothing reconciled the messages the candidate received against the record they described, which is what her personnel-file request was for and what the charge says it did not return. Nothing read the assessment result against five seasons of the same employer's own measurement of the same person. And nothing used what one automated speech-derived judgement got wrong to check the next one, four years apart, at the same employer.
assumed
Every other width comes from the PAN entry for this deployment on one stated rung mapping, applied to the eighteen pathways that have a one-to-one counterpart there with no exceptions. Three of the four PAN peer edges are redrawn as Lab checks, because a channel that improves practice is inhibiting in this vocabulary and reinforcing in PAN's; their widths still come from PAN. Five further PAN edges are folded into pathways this board already draws rather than drawn a second time, and the assumption below names every one of the five and says where its content lives. Every PAN edge for this deployment is marked estimated, so the drawn strengths are modelling choices in both files.
assumed
The standing workload reads as a bounded periodic cohort rather than a flood, because that is what the record describes: one internal promotion ladder at one employer, from tax associate to lead to seasonal manager, with leads trained toward the next step and many promoted each year, inside a seasonal business. No source publishes a per-cycle count of applicants for the role, and none is invented here.
baseline
The human comparator reads high, and the anchor is unusually direct. The counterfactual for this screen is not a hypothetical interview panel: it is five seasons of the same employer's own measurement of the same person — customer-satisfaction scores, resolution counts, response speed, annual bonuses every year, supervisor praise for her communication, a promotion already granted into a lead role over about four hundred associates, and a manager on the hiring team who had encouraged the application. That reading says the human baseline here is strong and documented. It does not say the screen is worse, and nothing on this board computes that: no head-to-head comparison exists anywhere in the record.
assumed
The assessed person is outside the network, as screened people are on every Lab diagram, and this deployment strains that convention in a way worth stating. She was simultaneously an incumbent employee, a member of the employer's own accessibility advisory body, and the person who had warned that body about this exact step. What is drawn is that the advisory channel existed and, as alleged, returned nothing. What is not drawn is her, and the pathway from an automated message to the person who reads it is narrated rather than drawn.
assumed
Five pathways drawn separately in this board's first derivation are folded into the survivors that already carry the same documented step, at the coarsest granularity at which every documented mechanism of this deployment is still distinguishable. The low script-adherence indicator reaching the selection decision is drawn once, along the performance record the charge itself says it sat in. The supervisors' continuous write into that record and the promotion round's own entry into it are properties of what the record holds, and they are stated on it. The single advisory warning is drawn once, on the channel that carried it, rather than again as a write against the assessment surface it concerned. And the accommodation correspondence's entry into the disclosable projection is stated on that projection and on the pathway the statutory right defines. Every fact each of the five carried is on this page; what changed is how many separate lines carry them.
baseline
No liability or wrongdoing by either respondent is asserted anywhere on this board, and no finding of discrimination is asserted by anyone against anyone. The charge is pending, the investigation is non-public, and both companies say the allegations are entirely without merit. Nothing here asserts that automated scoring ran on this interview, and nothing here asserts that it did not.

What this example does not show

Pending administrative charges: filed March 19, 2025 with the Colorado Civil Rights Division and the EEOC; investigation is non-public and no probable-cause determination, dismissal, right-to-sue notice, court filing, or settlement had surfaced as of 2026-08-28. Everything in this dossier from the charge is ALLEGATION, not finding. Both respondents dispute the charges on the record; HireVue disputes the central factual premise that an AI-based assessment was used at all.
The vendor's denial goes to the mechanism, not only to the law. Its chief executive states that the complaint rests on an inaccurate assumption about the technology used and that the employer did not use one of the vendor's artificial-intelligence-based assessments. The employer separately says the allegations are entirely without merit and that it provides reasonable accommodations to all candidates. Nothing on this board asserts that automated scoring ran on this interview, and nothing asserts that it did not.
The two pleaded layers have different evidentiary footings, and this board keeps them apart rather than blending them. The accommodation-denial sequence is alleged against the employer and survives either resolution of the technical dispute. The scoring allegation depends on that dispute and is pleaded on information and belief from a single message's return address and register.
The speech-recognition numbers belong to the study that made them. A peer-reviewed 2025 paper measured four widely used general-purpose commercial speech-recognition services on a twenty-four-speaker corpus: mean word error rate 52.6 percent for deaf and hard-of-hearing speakers against 5.0 percent for normal-hearing speakers, rising to 85.9 percent for the lowest speech-intelligibility group. It did not measure this platform, this interview or any component on this board, and the inference from that literature to this deployment is exactly what is contested.
The charge pleads only the Colorado Anti-Discrimination Act, the Americans with Disabilities Act and Title VII. Some secondary commentary mentions Colorado's 2024 artificial-intelligence statute as context; it is not a cause of action here and was not in effect at the events. Nothing on this board attributes an AI-statute claim to this charge.
The employer's call-monitoring metric is a different system from the promotion assessment, and the charge uses it as evidence of notice rather than as the screen that denied the promotion. That separation is carried on every pathway here: the two automated components share only a substrate and a check drawn at zero.
No number on this board measures either component. There is no published error rate, no accuracy figure and no independent evaluation of this deployment from any source, and every edge in the PAN entry for it is marked estimated. The absence is a finding of the case rather than a gap in the model: the assessed person could not establish from outside what had happened to her answers, and neither can anyone else.
The people this process sorted are outside the dynamics, as screened people are on every Lab diagram. No promotion outcome, score or rejection for any person is computed from anything drawn here. The assessed person is not named beyond the record's own initials and the roles it describes, and her first-person account was published on an ACLU affiliate site rather than as a statement of where she lives.
This board is the DEPLOYING EMPLOYER's internal promotion process. The vendor-layer board for the same interview platform, and the employer-liability board for another buyer of it, are different deployments with different evidence — their audits, their statutes, their settlements and their customer bases — and nothing from either is imported here.

Sources and evidence

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

On 19 March 2025 the ACLU, the ACLU of Colorado, Public Justice, and Eisenberg & Baum LLP filed a Complaint of Discrimination with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission on behalf of D.K., a Deaf Pawnee woman who communicates in English with a deaf accent and in ASL, against BOTH Intuit, Inc. and HireVue, Inc. It alleges violations of the Colorado Anti-Discrimination Act (Colo. Rev. Stat. § 24-34-402), the Americans with Disabilities Act and Title VII in the denial of her promotion to Seasonal Manager, and pleads HireVue as an employment agency, an agent of the employer, an indirect employer, and an aider and abettor under state law. None of those theories has been tested. Everything the charge asserts is an allegation and no probable-cause determination, dismissal, right-to-sue notice, court filing, or settlement has been made public.
empirical
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
- Trade press ACLU, ACLU of Colorado, Public Justice & Eisenberg & Baum LLP (2025, March). Civil-rights charges over an automated HireVue video interview at Intuit (Deaf and Indigenous applicant); via HR Dive and Fisher Phillips coverage. https://www.hrdive.com/news/ai-intuit-hirevue-deaf-indigenous-employee-discrimination-aclu/743273/
- Advocacy Public Justice (2025). Complaint Filed Against Intuit and HireVue Over Biased AI Hiring Technology That Violates Colorado Anti-Discrimination Act, ADA, and Title VII; URL returned HTTP 404 on 2026-08-28 and the release was not read https://www.publicjustice.net/hirevue-intuit-artificial-intelligence-biased-hiring/
This is an internal-promotion case rather than a point-of-hire case, and the employer held years of direct performance evidence about the person the screen was assessing. Per the complaint and D.K.'s own published account she was hired as a seasonal Tax Associate in late 2019, became a Tax Expert Lead supporting a team of about four hundred Tax Associates, held that role for three seasons, earned a bonus every year, was praised by supervisors for her communication, joined Intuit's Accessibility Team in 2023, and was encouraged to apply for Seasonal Manager by her own manager, who sat on the hiring team. The promotion ladder ran Tax Associate to Tax Expert Lead to Seasonal Manager, with Tax Expert Leads trained for the next step and, in the complaint's words, many being promoted each year. She applied in spring 2024 through the employee dashboard and was invited on 21 June 2024 into an assessment of roughly three hours: about a dozen timed recorded video questions, mostly management scenarios with a few tax-law questions, plus essay and multiple-choice sections, with instructions and questions delivered audibly by a recording of a person speaking.
empirical
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
- Advocacy ACLU of Oklahoma (2025, April 7). I Should Not Have to Fight for Fair Treatment in the Workplace (first-person account by the complainant) https://www.acluok.org/news/i-should-not-have-fight-fair-treatment-workplace/
The accommodation sequence as alleged: the interview invitation offered only a technical-support email address and carried no accommodations information, so D.K. initiated a request herself from prior familiarity with Intuit's process. She asked for Communication Access Real-time Translation (CART) — human-generated real-time captioning — deliberately not ASL interpretation, because interpreters she had worked with previously could not handle tax concepts, and she did not ask to be excused from the video interview because it was never offered and she feared that asking would count against her. The request was denied and she was told the platform's built-in subtitles could be enabled. When she began the assessment, she recounts, no subtitle option existed, and she completed roughly three hours of it relying on her browser's automatic captions, which she describes as less accurate and particularly poor on complicated words. The alleged failure therefore includes a control that was pointed to and was not present, rather than only a degraded substitute channel. Intuit's on-record position is that the allegations are entirely without merit and that it provides reasonable accommodations to all candidates.
empirical
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
- Advocacy ACLU of Oklahoma (2025, April 7). I Should Not Have to Fight for Fair Treatment in the Workplace (first-person account by the complainant) https://www.acluok.org/news/i-should-not-have-fight-fair-treatment-workplace/
- Trade press ACLU, ACLU of Colorado, Public Justice & Eisenberg & Baum LLP (2025, March). Civil-rights charges over an automated HireVue video interview at Intuit (Deaf and Indigenous applicant); via HR Dive and Fisher Phillips coverage. https://www.fisherphillips.com/en/insights/insights/ai-screening-systems-face-fresh-scrutiny-6-key-takeaways-from-claims-filed-against-hiring-technology-company
The complaint alleges Intuit was on notice twice over, through two channels of its own. First, after the 2023 season D.K. — by then a member of Intuit's Accessibility Team — told the Team's Chair that the HireVue step was inaccessible and could exclude deaf applicants; she was told it would be looked into, no action was ever reported back, and Intuit required the same step of her the following year. Second, in 2020 Intuit's OWN automated call-monitoring system, which scored how closely agents followed scripts from speech-recognition transcripts of customer calls, allegedly read her deaf accent as deviation from script and produced one artificially low indicator against otherwise strong customer-satisfaction, resolution-count, and response-speed measures. Intuit's alleged response was to reassign her from phone calls to the chat channel rather than to correct the metric; the low indicator remained in her record, and the complaint pleads it may also have been considered in the denial of her promotion. That call-monitoring system is Intuit's, not HireVue's, and the charge uses it as evidence of awareness rather than as the promotion screen.
empirical
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
- Advocacy ACLU of Oklahoma (2025, April 7). I Should Not Have to Fight for Fair Treatment in the Workplace (first-person account by the complainant) https://www.acluok.org/news/i-should-not-have-fight-fair-treatment-workplace/
On 13 August 2024 D.K. received an automated rejection stating that Intuit had decided to move forward with other candidates. On 25 September 2024 she received a feedback email — which she believed from its return address and generic language to have been generated by HireVue's automated analysis — advising her to give more concise and direct answers, to adapt her communication style to different audiences, and to practise active listening; the complaint says those recommendations map directly onto her being Deaf. On 31 October 2024 she made a personnel-file request under Colorado law, and the complaint filed the following March alleges that only minimal documents were produced. The complaint pleads HireVue's platform mechanism in a three-stage form — speech recognition of the spoken answers, a second system interpreting the transcript, and machine-learning scoring against job competencies — but pleads the scoring allegations about her specific interview on information and belief, inferred from that feedback email. That inference is exactly what HireVue disputes.
empirical
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
- Trade press Proskauer Rose LLP (2025, April 29). Another Legal Challenge to an AI Interviewing Tool; and Virginia Lawyers Weekly (2025, April 15). Deaf woman alleges AI bias in video interview process https://www.proskauer.com/blog/another-legal-challenge-to-an-ai-interviewing-tool
- Trade press ACLU, ACLU of Colorado, Public Justice & Eisenberg & Baum LLP (2025, March). Civil-rights charges over an automated HireVue video interview at Intuit (Deaf and Indigenous applicant); via HR Dive and Fisher Phillips coverage. https://www.hrdive.com/news/ai-intuit-hirevue-deaf-indigenous-employee-discrimination-aclu/743273/
The technical premise the charge builds on is a published measurement of general-purpose speech recognition, and it belongs to that study rather than to any deployment. A 2025 paper in The Laryngoscope (135:191-197) measured four widely used commercial speech-recognition services against a twenty-four-speaker corpus and reported a mean word error rate of 52.6 percent for d/Deaf and hard-of-hearing speakers versus 5.0 percent for normal-hearing speakers — per service 45.1 to 57.3 percent against 3.8 to 5.9 percent, which the authors describe as ten times higher — rising to 85.9 percent for speakers with the lowest speech-intelligibility classification, 80.5 percent for prelingual hearing-loss onset and 70.2 percent for sign-primary communicators. The complaint cites that study, together with peer-reviewed findings of higher speech-recognition error rates for Black speakers and for ethnicity-related dialects, and extends them on information and belief to speakers of Indigenous dialects of English. The study measured four general-purpose services. It did not measure HireVue's system or this interview, and the inference from that literature to this platform is contested.
empirical
- Academic Zhao, R., Choi, J. S., Koenecke, A., & Rameau, A. (2025). Quantification of Automatic Speech Recognition System Performance on d/Deaf and Hard of Hearing Speech. The Laryngoscope, 135, 191-197 https://koenecke.infosci.cornell.edu/files/Laryngoscope_Zhao2024.pdf
- Advocacy Complaint of Discrimination against Intuit, Inc. and HireVue, Inc. (redacted), filed 19 March 2025 with the Colorado Civil Rights Division and the U.S. Equal Employment Opportunity Commission by the ACLU, the ACLU of Colorado, Public Justice and Eisenberg & Baum LLP; read in full via the mirror hosted by the Law Office of Lainey Feingold because the filing organization's asset host returned HTTP 403 on 2026-08-28 https://www.lflegal.com/wp-content/uploads/2022/05/ACLU-complaint-against-HireVue-and-Intuit.pdf
Both respondents dispute the charges, and the vendor's dispute goes to the factual premise rather than to the legal conclusions. HireVue's chief executive Jeremy Friedman stated that the complaint 'is entirely without merit and is based on an inaccurate assumption about the technology used in the interview. Intuit did not use a HireVue AI-based assessment.' An Intuit spokesperson stated: 'The allegations in the complaint are entirely without merit. We provide reasonable accommodations to all candidates.' Whether any automated assessment ran on this interview at all is therefore in dispute, and the candidate-side evidence bearing on it is the return address and register of one feedback email. As of 28 August 2026 the matter remained a pending, non-public administrative investigation: no Colorado Civil Rights Division or EEOC determination, no right-to-sue notice, no court complaint, and no settlement had been publicly reported, and an April 2026 legal-commentary treatment of the case records no post-filing developments. Administrative charge proceedings are confidential, so that silence establishes no public development rather than that nothing occurred.
empirical
- Trade press ACLU, ACLU of Colorado, Public Justice & Eisenberg & Baum LLP (2025, March). Civil-rights charges over an automated HireVue video interview at Intuit (Deaf and Indigenous applicant); via HR Dive and Fisher Phillips coverage. https://www.hrdive.com/news/ai-intuit-hirevue-deaf-indigenous-employee-discrimination-aclu/743273/
- Academic Undergraduate Law Review at Florida State University (2026, April 29). Artificial Intelligence as a Hiring Tool: Hidden Discriminatory Impacts? https://ulrfsu.org/2026/04/29/artificial-intelligence-as-a-hiring-tool-hidden-discriminatory-impacts/
- Trade press ACLU, ACLU of Colorado, Public Justice & Eisenberg & Baum LLP (2025, March). Civil-rights charges over an automated HireVue video interview at Intuit (Deaf and Indigenous applicant); via HR Dive and Fisher Phillips coverage. https://www.fisherphillips.com/en/insights/insights/ai-screening-systems-face-fresh-scrutiny-6-key-takeaways-from-claims-filed-against-hiring-technology-company

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

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