The State of Play

AI-powered resume screening exemplifies the "good-practice" dilemma: mature, near-ubiquitous deployment (83–99% of enterprises) with proven efficiency gains, undermined by constitutional discrimination risks and regulatory liability that vendors and employers cannot yet mitigate at scale. The practice delivers measurable operational value—20–40% cost-per-hire reductions, 50–90% time-to-hire improvements—across healthcare, finance, and hospitality sectors. Yet independent peer-reviewed research across 2024–2026 confirms that all LLM-based and keyword-matching screeners systematically discriminate: University of Washington (ACM FAccT 2024) documented ChatGPT's explicit ableism against disability credentials; 2025 studies found 100% of tested LLMs exhibited gender bias and 85% white-name preference. Legal liability has crystallized: the Mobley v. Workday class action (EEOC-backed) covers potentially hundreds of millions rejected since 2020; Eightfold AI faces a novel FCRA lawsuit over undisclosed candidate scoring; and federal courts (March 2026) rejected dismissal motions on ADEA and disparate-impact claims. Regulatory fragmentation replaced federal guidance in early 2026—the Trump administration rescinded EEOC AI oversight, leaving Colorado AI Act, Illinois HB 3773, and state bias audit mandates as fragmented enforcement. The tier-defining tension: operational ubiquity and documented ROI constrained by systemic discrimination, unprecedented litigation, and regulatory patchwork that increases compliance burden without resolving fairness.

Deployment metrics show sustained operational value. HireVue reports 95% assessment completion rates, 92% candidate satisfaction, and over $667K in annual savings per enterprise customer; Children's Hospital of Philadelphia saved 1,695 hours annually by replacing manual phone screens. Unilever's video interview platform processed 250,000 applications with 90% time-to-hire reduction and 16% diversity improvement. Enterprise adoption has reached saturation: 99% of Fortune 500, 83% of all employers, and 87% of companies use AI in recruitment (2026 data). Production systems at this scale reflect consolidated vendor ecosystems—RChilli processes 4.1B+ documents annually across 1,600+ ERP platforms (Oracle HCM, SAP SuccessFactors, Salesforce); Eightfold AI serves major enterprises (Microsoft, Morgan Stanley, Starbucks). Yet deployment also reveals hidden failure modes: 19% of organizations using AI-only screening report overlooking qualified applicants; 40% of technical resumes are AI-optimized; candidate trust has eroded with only 26% of applicants trusting AI evaluation and 79% demanding transparency.

The bias and disability discrimination problem has shifted from academic evidence to enforceable legal and regulatory reality. LLM-based parsing, adopted as a supposed improvement over keyword matching, replicates discrimination at scale—all 22 LLMs tested in 2025 exhibited gender bias favouring female candidates; University of Washington (ACM FAccT 2024) found ChatGPT explicitly ranks resumes with disability credentials 75% lower; 85% of production screeners show white-name preference. Government agencies have responded: the US Department of Justice issued updated ADA guidance (March 2026) explicitly prohibiting hiring technologies that "unfairly screen out qualified individuals with disabilities," with employer liability even for third-party tool discrimination. Federal courts moved in parallel: in March 2026, a judge rejected Workday's dismissal motion on ADEA (age discrimination) claims, establishing that discrimination law covers entire hiring pipelines including AI screening. Eightfold AI now faces a novel FCRA class action (filed March 2026) alleging candidate scoring without disclosure, authorization, or dispute rights—expanding liability beyond employment law to consumer protection frameworks.

Regulatory enforcement shifted dramatically in early 2026. The Trump administration rescinded EEOC and DOL guidance on AI discrimination in January 2025, removing federal guardrails and legal clarity. States filled the vacuum: Colorado AI Act (effective February 2026), Illinois HB 3773 (January 2026), California regulations, and NYC bias audit mandates impose divergent compliance requirements on vendors and employers. This fragmentation creates a compliance barrier: employers using multi-state hiring cannot satisfy all requirements simultaneously, making deployment increasingly risky from a legal defensibility standpoint. Candidate sentiment and practitioner friction signal adoption resistance: only 26% of applicants trust AI evaluation; 49% of hiring managers auto-dismiss AI-generated resumes despite company reliance on screening; 40% of hiring teams cite bias as major concern; 83% of organizations lack AI maturity for reliable debiasing. The Mobley v. Workday class action, with EEOC backing for collective action certification covering potentially hundreds of millions rejected since 2020, remains pending and increasingly likely to establish employment-agency liability for major platforms.

• 2017: Enterprise adoption accelerating with Unilever (250k applicants, 4-month to 4-week cycle reduction) and HireVue deployments (Sabre, Hilton, Goldman Sachs). Bias risks surface: research shows anonymous screening doesn't eliminate age discrimination; Palantir settles discrimination lawsuit. Candidate experience barriers: 82% of job seekers frustrated with over-automation.

• 2018: Vendor consolidation and global expansion (pymetrics Series B $40M, HireVue 6M+ interviews across 700 customers). Amazon's AI screening tool exposed for systematic gender discrimination in October 2018 (penalizes women's language and colleges), becoming defining case of algorithmic bias; vendors commit to bias auditing. Deployment outcomes remain strong (Unilever: 75% time reduction, 20% diversity gain; Accenture, ANZ adoption). Academic advancement (ResumeNet research). Recruiter baseline: 7.4 seconds per resume, validating efficiency gains.

• 2019: Credibility crisis emerges as skepticism intensifies despite continued adoption (JPMorgan Chase, 700+ HireVue customers). Academic critics attack HireVue's facial analysis as pseudoscience; UC Berkeley analysis documents efficiency gains but flags cultural bias risks and algorithmic opacity. Candidate resistance grows: 76% of job seekers unwilling to apply to roles with computer screening. Research confirms screening as suitable AI deployment stage but highlights ROI and security concerns. Vendor positioning shifts toward behavioral science alternatives to resume-based assessment.

• 2020: Deployment scale accelerates despite regulatory escalation. HireVue wins US Intelligence Community $28.4M contract; Hitachi deploys for graduate hiring; Unilever maintains 75% efficiency and diversity gains. Academic R&D advances: University of Copenhagen secures DKK 7.1M for ML matching algorithms, researchers develop bias-mitigation techniques. However, legal and regulatory risks sharpen: Proskauer legal analysis warns of Title VII discrimination liability; US senators formally request EEOC investigation into algorithmic bias during COVID-19 hiring crisis. Practice maturity vs. accountability gap widens.

• 2021: Regulatory scrutiny accelerates and vendor responses diverge. HireVue removes facial expression analysis after independent ORCAA audit found nonverbal cues contributed negligible predictive power (0.25%) and introduced bias against minorities, signaling retreat from controversial screening methods. New York City passes AI hiring transparency law (effective 2023) requiring employer disclosure to candidates, foreshadowing litigation risks. Meanwhile, new vendors enter market (Circa, RChilli) with debiasing-focused offerings and expanded market accessibility. Academic evaluation reveals tool fragmentation: comparative analysis of leading commercial parsers (Textkernel, Joinvision, Sovren) shows significant variability in matching quality and parsing accuracy. By year-end, the practice shows divergent evolution: market maturity with scaled deployments persisting alongside regulatory acceleration, vendor feature retreat, and mounting legal uncertainty.

• 2022-H1: Regulatory intervention shifts from scrutiny to formal guidance; legal liability becomes concrete. EEOC issues first regulatory guidance on AI screening tool compliance under ADA (May 2022); class action lawsuit alleges HireVue collected biometric data without consent (January 2022), establishing BIPA liability. Independent academic research documents young job seekers' distrust of automation and shows socioeconomic privilege persists despite algorithmic screening. Vendor ecosystems expand (RChilli–Salesforce integration; HireVue's explainability statement) while market shows bifurcation: adoption continues at scale yet increasingly conditional on compliance auditing and candidate transparency.

• 2022-H2: Academic and regulatory evidence strengthens skeptical consensus. Peer-reviewed study of 694 recruiters finds algorithm aversion and behavioral risks when recommendation systems are inconsistent. Cambridge University study disputes vendor claims of bias reduction, calling video analysis pseudoscience. DOJ/EEOC guidance (October) clarifies ADA compliance and employer accountability for third-party tools. Deployment continues (Sitel: 35 avoided hires via screening), but compliance framework hardens: Canadian AI legislation proposed, NYC bias audit law looms (effective Jan 2023). By year-end, practice status: widespread deployment coupled with escalating legal, regulatory, and academic skepticism; adoption increasingly conditional on documented bias auditing and transparency mandates.

• 2023-H1: Regulatory and legal pressure escalates; public skepticism and practitioner doubts challenge automation hype. EEOC guidance extends to Title VII compliance (May 2023); lawsuits test vendor liability (Workday discrimination case, HireVue BIPA claims). Pew Research finds only 39% of Americans aware of AI in hiring and 71% oppose AI final decisions. HireVue's 2023 case studies show continued enterprise deployment (Flutter, Arm, Philips, Keurig Dr Pepper) with efficiency gains, yet practitioner analysis questions whether AI actually rejects candidates autonomously—most rejections still made by human recruiters. NYC bias audit law effective January 2023 accelerates demand for vendor transparency and algorithmic auditing. Practice bifurcates: enterprise scale and efficiency persist, yet regulatory compliance, public trust, and genuine automation maturity remain contested.

• 2023-H2: Enforcement action and system vulnerabilities expose practice fragility. EEOC wins first enforcement action against iTutorGroup for age discrimination in automated screening (August 2023, $365k settlement), validating discrimination risks documented in prior years. Security researchers reveal candidates circumvent AI screening via white-font keyword embedding, exposing pattern-matching simplicity and screening robustness gaps. Broader landscape: continued enterprise adoption and efficiency gains persist, yet enforcement actions, candidate gaming tactics, and regulatory compliance demands intensify focus on bias prevention, system transparency, and legal defensibility as prerequisites for sustained practice scale.

• 2024-Q1: Regulatory attention continues at supranational and practitioner levels; deployment and concern coexist. EU Parliament formalizes scrutiny of AI bias in recruitment tools tied to AI Act compliance. Enterprise case studies show continued efficiency gains: Hilton (50% time-to-fill reduction), Deloitte (60% screening time reduction with 30% diversity improvement), McDonald's (40% time-to-hire reduction). Practitioner surveys reveal bifurcation: 58.9% of hiring teams use resume screening tools yet 40% cite bias as major concern and 37% flag privacy risks. EEOC settlement with online tutoring company ($360k for age discrimination via programmed screening) reinforces legal liability. Vendor ecosystem matures: RChilli processes 4.1B+ documents annually with enterprise certifications (ISO 27001, SOC 2 Type II). Adoption accelerates but conditional on bias auditing and legal defensibility.

• 2024-Q2: Independent research confirms algorithmic bias risks; regulatory and legal pressure escalates. University of Washington study documents ChatGPT's explicit ableism in resume screening (disability credentials ranked lower 75% of the time). ACLU filed FTC complaint against Aon alleging discriminatory hiring AI despite 'bias free' marketing. EEOC filed amicus brief in Workday discrimination case arguing AI screening tools qualify as employment agencies under Title VII. California Civil Rights Department issued draft regulations restricting AI screening for adverse impact on protected characteristics. Positive deployment continues (healthcare: 200+ weekly screening tasks reduced from 45 min to minutes; $50K annual savings). Vendor innovation toward LLM-based parsing (RChilli beta). Yet evidence of systemic bias, legal liability, and regulatory intervention now documented across independent sources; practice faces growing tension between proven efficiency and unresolved fairness challenges.

• 2024-Q3: Regulatory and legal pressures intensify; adoption accelerates alongside risk awareness. Court allowed disparate impact claims to proceed against Workday (July 2024), signaling judicial recognition of systemic discrimination. Colorado, Illinois, and Utah AI laws finalize with 2025-2026 effective dates, adding compliance complexity. iHire survey shows AI recruitment adoption more than tripled year-over-year to 14.7% of employers (August 2024); HireVue survey reveals 79% of candidates demand transparency on AI use. Vendor ecosystem continues global expansion (RChilli adds Traditional Chinese support). Practitioner caution persists: 40% of hiring teams cite bias as major concern despite efficiency gains. Practice bifurcates further: proven deployment efficiency and rapid adoption scaling coexist with documented discrimination, escalating vendor liability, and regulatory intervention creating sustained uncertainty over bias mitigation and legal defensibility.

• 2024-Q4: Academic research documents persistent racial and gender bias in LLM-based screening; regulators issue compliance frameworks. University of Washington study (October 2024) auditing three LLMs found white-associated names preferred 85% of time, Black male names overlooked 100%, and Cornell Law review analyzing Amazon, Workday cases confirms systemic discrimination. Randomized controlled trial of 37,000 applicants (October 2024) showed AI-driven video interview screening achieved 20 percentage point improvement in pass rates vs traditional resume screening, though algorithm preferred younger candidates with less experience. UK Information Commissioner's Office issued audit recommendations (November 2024) highlighting bias and compliance risks in production AI recruitment tools; IAPP review documents 83% of US employers using automated hiring tools with state-level regulations (New York, Colorado, Illinois, Utah) taking effect 2025-2026. Workday discrimination lawsuit (Mobley case) progressed toward trial with EEOC support for Title VII employment agency classification. Vendor LLM evolution continues (RChilli integrations), but evidence base now clearly documents that generative AI approaches replicate historical training data biases as readily as keyword-matching systems. By year-end 2024, the practice stands at inflection: deployment efficiency proven and adoption accelerating (83% of employers), yet discrimination evidence, regulatory intervention, and legal liability frame continued scaling as contingent on bias mitigation and compliance infrastructure that remains unresolved.

• 2025-Q1: Deployments continue with efficiency outcomes; adoption accelerates amid regulatory patchwork and litigation escalation. Unilever case study (February 2025) reported 50,000 hours and £1M annual savings from AI video interview platform serving 250,000 applications with 90% time-to-hire reduction and 16% diversity increase. HireVue survey (February 2025) shows 61% of HR professionals implement AI in hiring (36% using resume screening), with weekly AI usage jumping to 72% in 2025 from 58% in 2024, signaling normalization. RChilli's GA resume parsing tool integrates with major ERP platforms (Oracle HCM, SAP SuccessFactors, Salesforce) claiming 85% drop-off reduction and 89% manual work reduction across 1,600+ platforms. However, Mobley v. Workday class-action received preliminary certification (February 2025) for nationwide collective action covering potentially 'hundreds of millions' harmed by AI screening discrimination, amplifying legal liability. Trump administration rescinded EEOC and DOL guidance on AI and discrimination (early 2025), removing federal guardrails, yet states intensified response: 27 state bills regulating AI hiring introduced in early 2025; NYC Local Law 144 bias audit requirements effective July 2023; Colorado AI Act (effective February 2026), Illinois HB 3773 (January 2026), Utah AI Policy, and California regulations impose compliance burdens. Critical analysis challenges vendor claims: ERE opinion (March 2025) argues AI resume screening fails to assess soft skills and replicates historical biases, citing Amazon's female discrimination case. By Q1 end, resume screening sits at a critical juncture: deployment efficiency and rapid adoption acceleration (72% weekly AI usage) coexist with federal deregulation, heightened state/local compliance requirements, pending nationwide class-action litigation with EEOC support, and mounting skepticism on bias mitigation and measurement validity.

• 2025-Q2: Academic research documents persistent algorithmic bias across all LLM architectures; collective-action litigation reaches nationwide scope; regulatory fragmentation intensifies. Two independent peer-reviewed studies published in May–June 2025 confirmed that LLM-based resume screening encodes systemic bias: University of Illinois study (May 2025) demonstrated hidden biases from resume variations like career gaps persist despite demographic redaction; City Journal study (June 2025) testing 22 LLMs found 100% exhibited gender bias favoring female candidates, contradicting vendor claims of AI-driven debiasing. Mobley v. Workday gained collective action certification (May 2025) for nationwide class covering potentially hundreds of millions of applicants rejected algorithmically since September 2020, with EEOC arguing the tool qualifies as employment agency under Title VII. Deployment efficiency gains continue: HireVue airline case (May 2025) shows skills assessment correlates with 3x customer acquisition and $2M annual value; Unilever video interview outcomes (February 2025) delivered 90% time-to-hire reduction and 16% diversity gains. Regulatory response bifurcated: EU AI Regulation classifies recruitment tools as high-risk (€35M fines); 27 US state bills introduced in 2025; Colorado, Illinois, California, NYC impose 2025–2026 effective dates with bias audit mandates. Vendor ecosystem continues LLM integration with major ERP platforms (Oracle, SAP, Salesforce), yet candidate trust remains fragile: 61% of job seekers believe AI reduces bias but 58% trust humans more; 31% used AI in job search (7-point increase from 2024). By Q2 end, the practice faces a credibility inflection: proven efficiency and 83% employer adoption coexist with conclusive evidence of discrimination across all LLM models, unprecedented class-action litigation, and regulatory patchwork that makes compliance and legal defensibility increasingly costly and uncertain.

• 2025-Q3: Academic bias research intensifies and expands; adoption accelerates despite discrimination evidence; vendor technical advancement continues. Brookings Institution research (August 2025) simulated LLM-mediated resume screening with demographically diverse name pool, confirming significant gender and racial discrimination including explicit bias against Black male candidates, echoing earlier findings and establishing bias as systematic rather than model-specific. Peer-reviewed research (September 2025) presents explainable NLP pipeline for resume parsing and job matching with 1000+ real resume dataset, advancing technical maturity for interpretable screening. Enterprise deployments continue with documented efficiency: Children's Hospital of Philadelphia achieved 1,695 hours saved annually and $667K cost savings via HireVue assessments eliminating manual phone screens. Resume.org survey (August 2025) of 1,399 US workers found 57% of companies already using AI in hiring, with 74% reporting improved hire quality and 1 in 3 anticipating full AI-driven hiring by 2026, indicating normalization and acceleration. Practitioner analysis (September 2025) comparing automated vs. human screening documents 95% accuracy for automated tools vs. 70% human accuracy and hybrid approach recommendations. By Q3 end, resume screening sits at a critical inflection: widespread deployment and proven efficiency gains coexist with independently-confirmed discrimination across LLM-based tools, mounting legal liability from class-action litigation with EEOC backing, and regulatory complexity spanning EU high-risk classification, state-level compliance mandates, and federal guidance vacuum created by Trump administration rollback of EEOC/DOL AI oversight. Bias mitigation remains unresolved—academic research confirms that generative AI screening replicates training data discrimination as readily as keyword-matching systems—yet adoption accelerates with third of employers planning full algorithmic hiring within 12 months.

• 2025-Q4: Category adoption reaches saturation; discrimination evidence crystallizes; legal liability escalates sharply. Adoption peaks: 88% of companies use AI candidate matching; 97.8% Fortune 500 employ ATS; 83% use AI specifically for resume screening with 81-96% reporting 1+ hour daily savings. HireVue production metrics show 95% completion, 92% satisfaction, $667K+ annual savings per customer. Yet discrimination evidence becomes definitive: ClassAction.org analysis (October 2025) documents 85% name bias across tools and 70% of companies allow AI to reject without human oversight; SIAI research (November 2025) confirms 83% adoption coupled with persistent 85% name preference despite debiasing claims; Compens.ai coverage (December 2025) reports 2025 LLM studies showing systematic bias against Black males and overpreference for females. Workday litigation escalates: Mobley v. Workday class-action covers potentially hundreds of millions rejected since September 2020; legal analysis (Kienbaum Hardy, November 2025) identifies employment-agency status liability and examines CVS HireVue settlement and Colorado disability case. Regulatory hardening: Colorado AI Act effective February 2026; Illinois HB 3773 effective January 2026; California, Utah, NYC impose bias audit mandates. Vendor evolution continues: Eightfold AI engineering blog (December 2025) details semantic matching and fairness design; RChilli integrates with SAP SuccessFactors (November 2025). By Q4 end, the practice exemplifies the good-practice inflection: near-universal enterprise adoption and documented efficiency gains coexist with conclusive peer-reviewed evidence of systemic discrimination, unprecedented class-action litigation with EEOC backing covering millions, and fragmented but rapidly hardening regulatory frameworks that increase compliance costs. The question for 2026: whether good-practice tier can hold while discrimination litigation, regulatory burden, and legal liability reshape vendor economics and adoption patterns.

• 2026-Jan: Legal and regulatory pressures crystallize; critical reassessment of AI effectiveness and bias emerges alongside sustained adoption. HBR expert analysis (January 2026) argues AI has worsened hiring through inefficiencies and algorithmic bias despite vendor claims. Adoption snapshot shows 80% of large companies using AI with 340% average ROI within 18 months (Taleva, January 2026), confirming continued deployment acceleration. Legal exposure expands: Mobley v. Workday class-action covers tens of millions with collective action certification; Eightfold AI faces FCRA violation lawsuit alleging undisclosed candidate ranking and rejection without human review—customers include Microsoft, Morgan Stanley, Starbucks. Vendor-critical analysis (January 2026) documents 85% white-name preference in screeners despite debiasing claims and advocates bespoke compliance-first tools. Legal expert guidance (JD Supra, January 2026) clarifies Title VII/ADA liability and regulatory frameworks (Colorado AI Act effective Feb 2026, Illinois HB 3773 effective Jan 2026, NYC Local Law 144 bias audits). By January end, the category shows bifurcation: adoption and ROI gains persist; yet expert consensus shifts toward skepticism on AI's ability to reduce bias at scale, legal liability escalates with second major lawsuit, and regulatory complexity increases compliance burden. Good-practice status remains but contingent on bias mitigation and legal defensibility measures that remain unproven.

• 2026-Feb: Regulatory framework activation and adoption paradoxes crystallize. Colorado AI Act effective February 2026 signals regulatory patchwork hardening; California AI hiring rules take effect adding compliance burden. Adoption remains near-saturation: 69% of HR professionals use AI recruiting, 83% of companies plan resume screening, 75% of large enterprises automate screening, but implementation shows friction: 19% of organizations report AI tools overlook qualified applicants, 49% of hiring managers auto-dismiss AI-generated resumes despite company reliance on AI screening, revealing candidate trust erosion and tool maturity gaps. Practitioner critical analysis intensifies: Josh Bersin industry report frames legal inflection as end of "Wild West" recruiting; DISHER Talent analysis cites 83% of organizations in lowest two AI maturity levels; only 26% of applicants trust AI evaluation despite 43% of companies using AI in HR. Talent leader adoption targets accelerate: 52% plan autonomous AI agent integration by 2026. By February end, the practice shows sustained deployment at scale coupled with mounting evidence of tool limitations, hiring manager skepticism, and regulatory compliance complexity. Efficiency gains (20-40% cost reduction) persist but candidate distrust and system maturity challenges threaten long-term adoption credibility.

• 2026-Mar: Legal exposure expanded on two new fronts: a federal court rejected Workday's ADEA dismissal motion (age discrimination claims now covering the full AI hiring pipeline), and Eightfold AI faces a novel FCRA class action for candidate scoring without disclosure or dispute rights — expanding liability beyond employment law into consumer protection. EEOC federal guidance removal (January 2025) left a fragmented state patchwork (Colorado, Illinois, California, NYC) that makes multi-state hiring compliance increasingly difficult. Brookings/UW research confirmed 85.1% of production screeners favour white-associated names, and DOJ issued updated ADA guidance explicitly prohibiting hiring technologies that unfairly screen out disabled candidates, establishing employer liability even for third-party tool discrimination.

• 2026-Apr: Adoption has reached near-saturation (82-83% employer deployment) but the system is exhibiting signal-corruption dynamics: 40% of candidates now use AI to draft resumes, 65% of hiring managers say AI-generated resumes slow hiring, and 75% of resumes are rejected by ATS before human review. A peer-reviewed causal fairness study (PopResume, 60.8K resumes, 4 LLMs and 4 VLMs) identified five distinct discrimination patterns that aggregate metrics fail to capture, while SHRM's survey of 1,908 HR professionals found the main barrier to scaling is governance and trust rather than technical capability.

• 2026-May: Adoption metrics solidify at extreme scale: Treegarden verified 98.8% Fortune 500 use ATS, 61% of recruiters use AI tools weekly, with EU AI Act enforcement deadline (August 2, 2026) and state regulatory frameworks (Colorado, Illinois, California, NYC) now active. Landmark litigation advances: federal court in Mobley v. Workday certified nationwide collective action under ADEA, with Workday's discovery disclosing 1.1 billion rejected applications and 23% higher rejection rates for applicants 40+. Eightfold AI faces novel FCRA litigation for undisclosed candidate scoring without disclosure or dispute rights, signaling vendor liability theory expansion beyond discrimination law. Independent research confirms critical control gap: SHRM survey of 1,908 HR professionals shows 80% use AI daily but 45% lack formal governance framework; Brookings/University of Washington study documents 85.1% white-name preference, with critical finding that human oversight alone is unreliable safeguard as recruiters mirror biased AI recommendations in ~90% of cases. Candidate experience deteriorating: 71% of applicants receive feedback from AI screening (vs. 31% silent rejection baseline), but only 26% trust AI evaluation and 69% express preference for human decision-making. Deployment stage remains production at massive scale, but governance maturity, legal defensibility, and fairness remains the inflection point determining whether established tier can hold against litigation and regulatory pressure.