Back

The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice

Emily BlackMiranda BogenLogan KoepkeSolon BarocasWesley DengMingwei Hsu

article2026Conference on Fairness, Accountability and Transparency0 citations

Reveals how decades of real-world fair lending compliance operate in practice, demonstrating through industry interviews that direct regulatory supervision—a mechanism largely missing from modern AI policy proposals—serves as the primary driver of algorithmic bias mitigation.

Listen

As policymakers propose new rules to govern artificial intelligence and mitigate automated bias, the challenge of translating high-level legal standards into effective corporate compliance has become urgent. U.S. financial institutions have operated fair lending compliance programs for decades under civil rights statutes such as the Equal Credit Opportunity Act and the Fair Housing Act, creating the longest-running operational testing ground for algorithmic fairness. The article evaluates how financial institutions test for and mitigate algorithmic discrimination in practice, examines how regulatory design influences these procedures, and identifies the practical and organizational challenges practitioners face in reducing disparities.

The authors conducted an empirical qualitative study based on 35 semi-structured interviews across the financial sector ecosystem. The participant pool included data science engineers, in-house and external lawyers, regulatory officials, and third-party compliance vendors. Using reflexive thematic analysis, the researchers developed an iterative codebook containing roughly 1,200 unique interpretation codes to analyze the workflows, institutional structures, and decision-making dynamics of fair lending programs.

The article yields four core findings. First, regulatory supervision drives compliance: proactive, routine supervisory examinations by regulators—rather than the threat of private lawsuits—compel financial institutions to maintain dedicated fair lending teams and standardized review processes that are largely absent in unregulated sectors. Second, while institutions universally conduct variable screening and disparate impact testing on consumer-facing models, specific metrics, thresholds, and methodologies vary widely due to an absence of granular regulatory standards. Third, organizational structures strictly separate first-line model developers from second-line compliance teams, restricting developers' access to demographic data and test results out of fear of triggering intentional discrimination (disparate treatment) claims; this separation creates operational friction and entrenches outdated bias mitigation techniques such as single-variable removal. Fourth, resolving identified disparities remains a commercial decision: business executives frequently reject less discriminatory models because firms tolerate virtually no performance degradation or profit loss (often requiring near-zero change in metrics like accuracy or area under the curve) and instead produce defensive documentation to justify disparities under business necessity standards.

These findings demonstrate that proactive regulatory oversight successfully establishes a baseline floor of fairness practices, but institutional incentives and perceived legal tensions cap substantive progress. Corporate risk calculations prioritize satisfying supervisory documentation requirements over achieving optimal demographic parity. Furthermore, modern artificial intelligence governance proposals that lack proactive supervisory mechanisms risk reproducing performative compliance without meaningful harm reduction.

Policymakers, regulators, and industry leaders should take concrete steps to improve algorithmic governance. Regulatory bodies should provide clearer, standardized guidance defining acceptable performance trade-offs and endorse modern race-aware algorithmic mitigation techniques that do not trigger disparate treatment penalties. AI governance frameworks in other sectors must incorporate active supervisory examination models rather than relying solely on reactive, complaint-driven enforcement. Organizations should also streamline workflows between model developers and compliance teams to address disparities earlier during model design.

The study's primary limitations stem from its qualitative sample, which did not include representatives from small financial institutions or state-level regulatory agencies. In addition, legal confidentiality and ongoing shifts in federal enforcement policy may influence participant perspectives. Nonetheless, the findings offer high-confidence empirical insights into how regulatory design directly shapes the effectiveness of algorithmic fairness programs on the ground.

Carnegie Mellon UniversityCenter for Democracy & TechnologyMicrosoftNew York UniversityUpturnpurchasebusiness processincentive mechanismFair Lending Modelalgorithmic fairness programsfairness testing
arXiv: 2606.02957
  • Paper: Big Data's Disparate Impact, Solon Barocas et al. (2016). It provides the foundational legal and technical analysis of how algorithmic data mining produces disparate impact under U.S. civil rights law, which directly underpins the fair lending compliance models analyzed in the source.
  • Paper: Fairness and Abstraction in Sociotechnical Systems, Andrew D. Selbst et al. (2019). It frames how technical fairness abstractions fail in real-world institutional contexts, establishing the sociotechnical perspective that the source investigates empirically through fair lending compliance workflows.
  • Paper: Certifying and Removing Disparate Impact, Michael Feldman et al. (2014). It establishes key methodology for certifying and mitigating statistical disparate impact in predictive decision models, providing the technical basis for algorithmic testing programs evaluated in the source.
  • Paper: Fairness Constraints: Mechanisms for Fair Classification, Muhammad Bilal Zafar et al. (2015). It formulates mathematical mechanisms to balance accuracy with legal disparate impact and business necessity constraints, mirroring the exact tensions faced by lending practitioners.
  • Paper: Equality of Opportunity in Supervised Learning, Moritz Hardt et al. (2016). It introduces formal statistical criteria such as equal opportunity and equalized odds in supervised learning, establishing standard benchmarks used by institutions to test algorithmic bias.
  • Paper: Delayed Impact of Fair Machine Learning, Lydia T. Liu et al. (2018). It models the downstream feedback loops of fair lending constraints over time using credit data, providing theoretical context for the practical trade-offs encountered in fair lending programs.
  • Paper: Inherent Trade-Offs in the Fair Determination of Risk Scores, Jon Kleinberg et al. (2017). It proves the mathematical impossibility of simultaneously satisfying competing fairness definitions in risk scoring, elucidating the core regulatory uncertainties and technical dilemmas described in the source.
  • Paper: Fairness through awareness, Cynthia Dwork et al. (2012). It defines seminal principles for individual and group fairness in high-stakes classification settings like credit underwriting.

No sufficiently relevant recommendations were found.

Cover for The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice

Abstract

U.S. financial institutions subject to fair lending laws have been running algorithmic fairness programs for decades. Despite this long history, remarkably little is known about how these requirements operate in practice. In this paper, we offer the first empirical account of how financial institutions test for and mitigate algorithmic discrimination on the ground. In doing so, we shed light on how the regulatory design of fair lending law and regulation have shaped the policies, processes, and practices of fair lending programs. Drawing on 35 semi-structured interviews with participants across the fair lending ecosystem, we find that while financial institutions have a floor of fairness practices aimed at preventing discrimination in lending largely absent in other domains, the specifics of how firms test for discrimination and search for less discriminatory algorithms varies widely. We also find that regulatory supervision via fair lending examinations has been the key driver of compliance work, but that the practical impact of fair lending programs often depends on how well they can navigate competing business incentives, perceived legal tensions, and regulatory uncertainty. Ultimately, our findings highlight the unique role that supervisory authority has played in successfully fostering fair lending practices -- a regulatory design feature that is distinct from other areas of civil rights law and almost completely absent from recent policy proposals for dealing with algorithmic discrimination.

Topics

purchasebusiness processincentive mechanismdesign of algorithmsproportional fairness

Keywords

Fair Lending Modelalgorithmic fairness programsfairness testingFairness Toolkitsalgorithmic biasalgorithmic discriminationFair Housing ActEqual Credit Opportunity Actbias auditbias mitigation

Table of Contents

  • 1 Introduction
  • 2 Background
  • 3 Related Work
  • 4 Methods
  • 5 Findings
  • 5.1 RQ1: The contours of fair lending practice
  • 5.1.1 Existence of Fair Lending Teams and Practices
  • 5.1.2 Procedures to Test for Disparate Treatment and Disparate Impact
  • 5.1.3 Searching for Less Discriminatory Algorithms.
  • 5.2 RQ2: Driving forces of discrimination work in regulated financial institutions
  • 5.2.1 Participants cite regulatory expectations as a reason why their work was done at all.
  • 5.2.2 Supervisory expectations and fair lending examinations as a primary motivator
  • 5.3 RQ3: Challenges to fair lending work in practice
  • 5.3.1 Frustration with compliance being a business decision
  • 5.3.2 Perceived tensions: disparate treatment and impact
  • 5.3.3 Navigating uncertainty around regulatory expectations, and its consequences
  • 6 Discussion
  • 6.1 More mature practices are still plagued by familiar challenges
  • 6.2 The mixed success of fair lending offers useful lessons for regulatory proposals for AI governance
  • 6.3 Discretion is a double-edged sword
  • 7 Limitations
  • 8 Conclusion
  • 9 Generative AI Statement
  • References
  • A Participant Information
  • B Interview Protocols
  • B.1 Policymakers
  • B.2 Engineers
  • B.3 Regulators
  • C Codebook

Knowls

  1. Knowl 1 — Supervisory Examination as the Dominant Driver of Algorithmic Antidiscrimination Compliance

    empirical result

    In U.S. consumer financial services, algorithmic fairness practices are almost exclusively driven by proactive regulatory supervision and routine fair lending examinations conducted by financial regulators (such as the Consumer Financial Protection Bureau), rather than by private civil litigation or voluntary corporate AI ethics commitments. Regulated financial entities establish dedicated fair lending compliance units and formal algorithmic review protocols because examinations occur regularly and non-compliance carries severe regulatory consequences. Conversely, in sectors where routine supervisory examination is absent (such as insurance) or for protected classes that examiners do not actively probe (such as disability or national origin, despite statutory coverage under fair lending law), firms rarely perform algorithmic disparity testing. Practitioners note that testing algorithms for bias in the absence of a mandatory supervisory requirement is frequently perceived by corporate legal teams as creating unshielded legal risk.

  2. Knowl 2 — Organizational Firewall Between Model Developers and Compliance Teams

    empirical result

    To mitigate perceived legal risks under disparate treatment doctrine—which prohibits making lending decisions based on protected characteristics—financial institutions implement a strict institutional firewall between model developers (the "first line of defense") and fair lending compliance teams (the "second line of defense"). Under this separation, model developers are denied access to demographic data and are frequently barred from seeing the quantitative results of disparate impact analyses performed by compliance teams. When a model fails a fairness audit, developers are instructed to modify the model without access to the disparity breakdown, creating an inefficient trial-and-error cycle ("throwing models over the wall"). While this firewall was manageable for simpler regression models, practitioners report it creates a severe barrier to developing and optimizing modern machine learning systems. Some institutions have begun testing restricted automated bias estimators that enforce query budgets to prevent developers from indirectly reverse-engineering demographic variables.

  3. Knowl 3 — Business Priority and Asymmetric Performance Degradation Tolerances in Less Discriminatory Algorithm Selection

    empirical result

    Within financial institutions, the authority to approve or reject a Less Discriminatory Algorithm (LDA)—an alternative model that achieves comparable performance with reduced demographic disparity—resides with business units rather than fair lending teams. Under disparate impact doctrine in the Equal Credit Opportunity Act (ECOA), institutions are permitted to maintain a disparate model if they can document a "legitimate business need." While firms regularly accept model performance degradation of 2.5%2.5\% to 5.0%5.0\% to accommodate data drift under safety and soundness model risk management, business units enforce an asymmetric, near-zero performance loss tolerance for fair lending alternatives (e.g., rejecting an alternative credit model if its Kolmogorov-Smirnov (KS) statistic drops from 0.500.50 to 0.480.48 or 0.490.49, or if the Area Under the ROC Curve (AUC) decreases by 10−610^{-6}). As a result, candidate LDAs that reduce outcome disparities are routinely dismissed on business profitability grounds.

  4. Knowl 4 — Prevalence of Drop-One Feature Ablation and Defensive Justification Documentation

    empirical result

    Rather than adopting modern in-processing debiasing methods from the algorithmic fairness literature (such as adversarial debiasing or constrained optimization), financial institutions predominantly rely on legacy "drop-one" analysis—systematically removing one input feature at a time and retraining the model—to search for Less Discriminatory Algorithms (LDAs). More advanced debiasing techniques that incorporate protected attribute data during model training are widely rejected by institutional counsel due to the fear that using demographic data in training constitutes unlawful disparate treatment. When viable LDAs cannot be found via drop-one analysis or are rejected for performance degradation, fair lending efforts shift toward creating "defensive documentation" (such as business justification reports or memos). These artifacts detail the model architecture, feature sets, and statistical validation to justify to regulatory examiners that remaining demographic disparities reflect legitimate, baseline credit risk rather than avoidable discrimination.

  5. Knowl 5 — Tiered Algorithmic Disparity Testing Pipeline in Fair Lending Compliance

    model/method

    Fair lending compliance teams evaluate credit scoring and underwriting algorithms using a structured, tiered pipeline:

    1. Model Risk Triage: Models are categorized by risk level based on consumer proximity; consumer-facing underwriting and pricing models receive comprehensive disparate impact analysis, while lower-impact internal models receive abbreviated assessments.
    2. Feature Review (Disparate Treatment Screening): Candidate model features undergo qualitative and quantitative vetting using categorized variable lists (green for clear credit relationship, yellow requiring further statistical scrutiny, and red for prohibited attributes or direct proxies) to ensure no explicit prohibited bases or close proxies enter the feature set.
    3. Disparate Impact Testing: Risk-scored models are evaluated for outcome disparities across demographic groups using statistical metrics such as Adverse Impact Ratios (AIR), Standardized Mean Differences (SMD), and marginal effects. In some pipelines, predicted default distributions across demographic classes are benchmarked against historical default base rates; if the predicted disparity does not worsen historical default rate differentials, the model is classified as carrying low fair lending risk.

  6. Knowl 6 — Bayesian Improved Surname Geocoding (BISG) as the De Facto Industry Standard for Demographic Imputation

    empirical result

    Because financial institutions generally do not collect self-reported race and ethnicity data for non-mortgage credit applications, firms rely on Bayesian Improved Surname Geocoding (BISG) to proxy applicant race and ethnicity for algorithmic disparity testing. The widespread adoption of BISG was catalyzed by the Consumer Financial Protection Bureau (CFPB) publishing a whitepaper validating its methodology, which provided regulated entities with a standardized, regulator-accepted protocol. However, the availability and regulatory acceptance of BISG has created a disincentive for lenders to collect actual self-reported demographic data, while leaving institutions without standardized proxying tools for other legally protected categories under ECOA, such as sex, national origin, and disability.

  7. Knowl 7 — The Multi-Group 'Train-Crash' Challenge in Empirical Fair Lending

    empirical result

    In production lending environments, fair lending testing requires evaluating model behavior across a multidimensional demographic matrix (described by practitioners as a "2 by 7 matrix" spanning combinations of protected attributes such as race, sex, age, and marital status). Practitioners report an operational challenge termed the "train-crash" problem: modifying model features, loss weights, or decision thresholds to mitigate disparities for one protected subpopulation frequently causes an increase in disparities for another protected group. Although theoretical multi-group fairness and multicalibration frameworks exist in the literature, practitioners lack usable, production-ready tooling that can resolve combinatorial trade-offs while satisfying legal compliance standards.

  8. Knowl 8 — Semi-Structured Interview Study Design on Fair Lending Compliance Practices

    experimental setup

    The empirical study examined algorithmic fairness and compliance practices within the U.S. financial services domain by conducting 35 semi-structured interviews (totaling approximately 38 hours of audio) with professionals across four primary populations:

    • Engineers (EE, n=10n = 10)
    • Lawyers (LL, n=7n = 7)
    • Third-party vendors and consultants (TT, n=10n = 10)
    • Financial regulators (RR, n=8n = 8)

    Participants were recruited through purposive and snowball sampling from over 100 contacted practitioners across commercial banks, regulatory agencies, and technology vendors. Transcripts were analyzed using reflexive thematic analysis, generating approximately 1,200 interpretation codes categorized across eight high-level thematic dimensions: Policies (rules), Procedures, Processes (practice/standards), Regulation from Entities' Point of View, Documentation, Regulation from Regulators' Point of View, Organizational Dynamics, and Financial Costs.

  9. Knowl 9 — Fragility of Discretion-Based Supervisory Regimes Under Political Turnover

    theoretical result

    Regulatory models that rely primarily on agency supervisory discretion and examination rather than bright-line statutory mandates create robust compliance incentives during active enforcement periods, but remain inherently vulnerable to political turnover across government administrations. Because supervisory priorities and examination scopes can be altered administratively without formal legislative action or notice-and-comment rulemaking (such as administrative directives halting disparate impact enforcement or statistical bias examinations under ECOA Regulation B), supervisory-driven algorithmic accountability lacks long-term institutional stability if it is not codified into durable statutory requirements.

  10. Knowl 10 — Sampling and Observational Scope Limitations

    limitation

    The findings are subject to several constraints:

    1. Sample Composition: The 35 participants represent large financial institutions, federal regulatory agencies, and major industry vendors, excluding smaller community banks, credit unions, and state-level financial regulators.
    2. Social Desirability and Non-Disclosure Constraints: Due to strict legal exposure and corporate non-disclosure requirements surrounding fair lending audits, participant accounts may be subject to social desirability bias or omit confidential internal compliance practices.
    3. Regulatory Fluidity: The empirical data captures organizational practices immediately preceding federal administrative shifts that altered disparate impact enforcement posture under Regulation B.

Coverage note — None was omitted; all substantive findings, empirical data, methodology, and structural analyses from the paper have been converted into self-contained knowls.

References

  1. 1.
    1. Fair Housing Act, 42 U.S.C. §§ 3601–3619.
  2. 2.Alekh Agarwal, Alina Beygelzimer, Miroslav Dudík, John Langford, and Hanna Wallach. 2018. A Reductions Approach to Fair Classification. In Proceedings of the 35th International Conference on Machine Learning (ICML). PMLR, 60–69. Introduces a framework for enforcing fairness constraints via reductions to cost-sensitive classification.
  3. 3.Muhammad Azeem Akbar, Arif Ali Khan, Sajjad Mahmood, Saima Rafi, and Selina Demi. 2024. Trustworthy artificial intelligence: A decision-making taxonomy of potential challenges. Software: Practice and Experience 54, 9 (2024), 1621–1650.
  4. 4.Pouria Akbarighatar. 2024. Operationalizing responsible AI principles through responsible AI capabilities. AI and Ethics (2024), 1–15.
  5. 5.Sanna J Ali, Angèle Christin, Andrew Smart, and Riitta Katila. 2023. Walking the walk of AI ethics: Organizational challenges and the individualization of risk among ethics entrepreneurs. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency. 217–226.
  6. 6.McKane Andrus, Elena Spitzer, Jeffrey Brown, and Alice Xiang. 2021. What we can’t measure, we can’t understand: Challenges to demographic data procurement in the pursuit of fairness. In Proceedings of the 2021 ACM conference on fairness, accountability, and transparency. 249–260.
  7. 7.Agathe Balayn, Mireia Yurrita, Jie Yang, and Ujwal Gadiraju. 2023. “Fairness Toolkits, A Checkbox Culture?” On the Factors that Fragment Developer Practices in Handling Algorithmic Harms. In Proceedings of the 2023 AAAI/ACM Conference on AI, Ethics, and Society. 482–495.
  8. 8.Robert Baldwin, Martin Cave, and Martin Lodge. 2011. Understanding regulation: theory, strategy, and practice. Oxford university press.
  9. 9.Kenneth A Bamberger and Deirdre K Mulligan. 2015. Privacy on the ground: driving corporate behavior in the United States and Europe. MIT Press.
  10. 10.Rebecca Bauer-Kahan. 2025. Automated Decision Systems (Automated Decisions Safety Act). California State Legislature. https://leginfo.legislature.ca.gov/faces/billNavClient.xhtml?bill_id=202520260AB1018 A.B. 1018, 2025–2026 Regular Session (Cal.).
  11. 11.Emily Black et al. 2024. Less Discriminatory Algorithms. Georgetown Law Journal 113 (2024), 53–.
  12. 12.Board of Governors of the Federal Reserve System. 2007. Report to the Congress on Credit Scoring and Its Effects on the Availability and Affordability of Credit. Technical Report. Federal Reserve Board. https://www.federalreserve.gov/boarddocs/rptcongress/creditscore/ Submitted pursuant to section 215 of the Fair and Accurate Credit Transactions Act of 2003.
  13. 13.Miranda Bogen, Aaron Rieke, and Shazeda Ahmed. 2020. Awareness in practice: tensions in access to sensitive attribute data for antidiscrimination. In Proceedings of the 2020 Conference on Fairness, Accountability, and Transparency (Barcelona, Spain) (FAT* ’20). Association for Computing Machinery, New York, NY, USA, 492–500. doi:10.1145/3351095.3372877
  14. 14.Rishi Bommasani, Sanjeev Arora, Jennifer Chayes, Yejin Choi, Mariano-Florentino Cuéllar, Li Fei-Fei, Daniel E. Ho, Dan Jurafsky, Sanmi Koyejo, Hima Lakkaraju, Arvind Narayanan, Alondra Nelson, Emma Pierson, Joelle Pineau, Scott Singer, Gaël Varoquaux, Suresh Venkatasubramanian, Ion Stoica, Percy Liang, and Dawn Song. 2025. Advancing science- and evidence-based AI policy. Science 389, 6759 (31 July 2025), 459–461. doi:10.1126/science.adu8449 Policy Forum.
  15. 15.Virginia Braun and Victoria Clarke. 2019. Reflecting on reflexive thematic analysis. Qualitative Research in Sport, Exercise and Health 11, 4 (2019), 589–597. doi:10.1080/2159676X.2019.1628806 Paper discussing developments and clarifications of reflexive thematic analysis.
  16. 16.Virginia Braun and Victoria Clarke. 2021. Thematic analysis: A practical guide. (2021).
  17. 17.Jacques Bughin. 2024. Doing versus saying: responsible AI among large firms. AI & SOCIETY (2024), 1–13.
  18. 18.California State Legislature. 2025. Transparency in Frontier Artificial Intelligence Act, SB 53. State Law.
  19. 19.Noel Capon. 1976. Credit Scoring Systems: A Critical Analysis. Journal of Marketing Research (1976). https://business.columbia.edu/sites/default/files-efs/pubfiles/690/24.pdf Columbia Business School Working Paper version.
  20. 20.Simon Caton and Christian Haas. 2024. Fairness in Machine Learning: A Survey. Comput. Surveys 56, 7 (2024), 1–38. doi:10.1145/3616865 Comprehensive overview of fairness definitions and methods, organizing pre-processing, in-processing, and post-processing approaches.
  21. 21.Alexandra Chouldechova. 2017. Fair prediction with disparate impact: A study of bias in recidivism prediction instruments. Big data 5, 2 (2017), 153–163.
  22. 22.Robert Cinca, Enrico Costanza, and Mirco Musolesi. 2025. Practitioners and Bias in Machine Learning: A Study. ACM Transactions on Interactive Intelligent Systems 15, 2 (2025), 1–28.
  23. 23.Julie E. Cohen, Nina-Simone Edwards, Meg Leta Jones, and Paul Ohm. 2025. Designing Policymaking Mechanisms for Regulatory Dynamism. Preliminary Concept Paper. Georgetown University Law Center, Institute for Technology Law & Policy, Reimagining the Governance Stack Project.
  24. 24.Julie E Cohen, Paul Ohm, Meg Leta Jones, Brenda Dvoskin, and Smitha Krishna Prasad. 2024. Regulatory Monitoring in the Information Economy. Redesigning the Governance Stack Project (2024).
  25. 25.Colorado State Legislature. 2024. Colorado Anti-Discrimination in AI Law / Consumer Protections for Artificial Intelligence (Colorado AI Act), SB 24-205. State Law.
  26. 26.Consumer Financial Protection Bureau. 2014. Using Publicly Available Information to Proxy for Unidentified Race and Ethnicity. White Paper. https://files.consumerfinance.gov/f/201409_cfpb_report_proxy-methodology.pdf
  27. 27.Consumer Financial Protection Bureau. 2025. Equal Credit Opportunity Act (Regulation B). Federal Register 90 (13 Nov. 2025), 50901–50923. Proposed Rule, Document No. 2025-19864, RIN 3170-AB54.
  28. 28.Peter Conti-Brown and Sean H. Vanatta. 2025. Private Finance, Public Power: A History of Bank Supervision in America. Princeton University Press.
  29. 29.Sasha Costanza-Chock, Inioluwa Deborah Raji, and Joy Buolamwini. 2022. Who Audits the Auditors? Recommendations from a field scan of the algorithmic auditing ecosystem. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 1571–1583.
  30. 30.Wesley Hanwen Deng, Solon Barocas, and Jennifer Wortman Vaughan. 2025. Supporting Industry Computing Researchers in Assessing, Articulating, and Addressing the Potential Negative Societal Impact of Their Work. Proceedings of the ACM on Human-Computer Interaction 9, 2 (2025), 1–37.
  31. 31.Wesley Hanwen Deng, Boyuan Guo, Alicia Devrio, Hong Shen, Motahhare Eslami, and Kenneth Holstein. 2023. Understanding practices, challenges, and opportunities for user-engaged algorithm auditing in industry practice. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–18.
  32. 32.Wesley Hanwen Deng, Manish Nagireddy, Michelle Seng Ah Lee, Jatinder Singh, Zhiwei Steven Wu, Kenneth Holstein, and Haiyi Zhu. 2022. Exploring how machine learning practitioners (try to) use fairness toolkits. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 473–484.
  33. 33.Wesley Hanwen Deng, Nur Yildirim, Monica Chang, Motahhare Eslami, Kenneth Holstein, and Michael Madaio. 2023. Investigating practices and opportunities for cross-functional collaboration around AI fairness in industry practice. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency. 705–716.
  34. 34.Cynthia Dwork, Moritz Hardt, Toniann Pitassi, Omer Reingold, and Richard Zemel. 2012. Fairness through awareness. In Proceedings of the 3rd innovations in theoretical computer science conference. 214–226.
  35. 35.Lauren B Edelman. 2020. Working law: Courts, corporations, and symbolic civil rights. University of Chicago Press.
  36. 36.Equal Employment Opportunity Commission. 1983. Procedures for Complaints of Employment Discrimination Filed Against Recipients of Federal Financial Assistance. 29 C.F.R. Part 1691. Originally published 48 FR 3574 (Jan. 25, 1983); last amended 54 FR 32063 (Aug. 4, 1989). Electronic Code of Federal Regulations, current as of Apr. 30, 2026. https://www.ecfr.gov/current/title-29/subtitle-B/chapter-XIV/part-1691.
  37. 37.FinRegLab. 2023. Explainability and Fairness in Machine Learning for Credit Underwriting: Policy Analysis. https://finreglab.org/research/explainability-fairness-in-machine-learning-for-credit-underwriting-policyanalysis/ Dec. 2023.
  38. 38.FinRegLab. 2023. Machine Learning Explainability & Fairness: Insights from Consumer Lending. https://finreglab.org/research/machine-learning-explainability-fairness-insights-from-consumer-lending/ Jul. 2023.
  39. 39.Marissa Kumar Gerchick, Ro Encarnación, Cole Tanigawa-Lau, Lena Armstrong, Ana Gutiérrez, and Danaé Metaxa. 2025. Auditing the Audits: Lessons for Algorithmic Accountability from Local Law 144’s Bias Audits. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency (FAccT ’25). Association for Computing Machinery, New York, NY, USA, 29–44. doi:10.1145/3715275.3732004
  40. 40.Jessica Gonzalez. 2025. New York Artificial Intelligence Consumer Protection Act. New York State Senate. https://www.nysenate.gov/legislation/bills/2025/S1962 S. 1962, 2025–2026 Regular Sessions (N.Y.)..
  41. 41.Parikshit Gopalan, Adam Tauman Kalai, Omer Reingold, Vatsal Sharan, and Udi Wieder. 2021. Omnipredictors. arXiv preprint arXiv:2109.05389 (2021).
  42. 42.Lara Groves, Jacob Metcalf, Alayna Kennedy, Briana Vecchione, and Andrew Strait. 2024. Auditing Work: Exploring the New York City algorithmic bias audit regime. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). Association for Computing Machinery, New York, NY, USA, 1107–1120. doi:10.1145/3630106.3658959
  43. 43.Neel Guha, Christie M. Lawrence, Lindsey A. Gailmard, Kit T. Rodolfa, Faiz Surani, Rishi Bommasani, Inioluwa Deborah Raji, Mariano-Florentino Cuéllar, Colleen Honigsberg, Percy Liang, and Daniel E. Ho. 2024. AI Regulation Has Its Own Alignment Problem: The Technical and Institutional Feasibility of Disclosure, Registration, Licensing, and Auditing. The George Washington Law Review 92, 6 (Dec. 2024), 1473–1557. https://www.gwlr.org/ai-regulation-has-its-own-alignment-problem-the-technical-and-institutional-feasibility-of-disclosure-registration-licensing-and-auditing/
  44. 44.Lakshitha Gunasekara, Nicole El-Haber, Swati Nagpal, Harsha Moraliyage, Zafar Issadeen, Milos Manic, and Daswin De Silva. 2025. A Systematic Review of Responsible Artificial Intelligence Principles and Practice. Applied System Innovation 8, 4 (2025), 97.
  45. 45.Patrick Hall, Benjamin Cox, Steven Dickerson, Arjun Ravi Kannan, Raghu Kulkarni, and Nicholas Schmidt. 2021. A United States Fair Lending Perspective on Machine Learning. Frontiers in Artificial Intelligence 4 (June 2021), 695301. doi:10.3389/frai.2021.695301 Mini Review article, Section: AI in Finance.
  46. 46.Moritz Hardt, Eric Price, and Nathan Srebro. 2016. Equality of Opportunity in Supervised Learning. In Advances in Neural Information Processing Systems, Vol. 29. NeurIPS, 3315–3323. Proposes the “equal opportunity” fairness criterion for predictors in supervised learning.
  47. 47.Ursula Hébert-Johnson, Michael Kim, Omer Reingold, and Guy Rothblum. 2018. Multicalibration: Calibration for the (computationally-identifiable) masses. In International Conference on Machine Learning. PMLR, 1939–1948.
  48. 48.Daniel E Ho and Alice Xiang. 2020. Affirmative algorithms: The legal grounds for fairness as awareness. U. Chi. L. Rev. Online (2020), 134.
  49. 49.Kenneth Holstein, Jennifer Wortman Vaughan, Hal Daumé III, Miro Dudik, and Hanna Wallach. 2019. Improving fairness in machine learning systems: What do industry practitioners need?. In Proceedings of the 2019 CHI conference on human factors in computing systems. 1–16.
  50. 50.Pauline T. Kim. 2022. Race-Aware Algorithms: Fairness, Nondiscrimination, and Affirmative Action. California Law Review 110 (2022), 1539–.
  51. 51.I. Elizabeth Kumar, Keegan E. Hines, and John P. Dickerson. 2022. Equalizing Credit Opportunity in Algorithms: Aligning Algorithmic Fairness Research with U.S. Fair Lending Regulation. In Proceedings of the 2022 AAAI/ACM Conference on AI, Ethics, and Society (Oxford, United Kingdom) (AIES ’22). Association for Computing Machinery, New York, NY, USA, 357–368. doi:10.1145/3514094.3534154
  52. 52.Khoa Lam, Benjamin Lange, Borhane Blili-Hamelin, Jovana Davidovic, Shea Brown, and Ali Hasan. 2024. A Framework for Assurance Audits of Algorithmic Systems. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). ACM, New York, NY, USA, 1480–1494. doi:10.1145/3630106.3658957
  53. 53.Hao-Ping Hank Lee, Lan Gao, Stephanie Yang, Jodi Forlizzi, and Sauvik Das. 2024. "I Don’t Know If We’re Doing Good. I Don’t Know If We’re Doing Bad": Investigating How Practitioners Scope, Motivate, and Conduct Privacy Work When Developing {AI} Products. In 33rd USENIX Security Symposium (USENIX Security 24). 4873–4890.
  54. 54.Michael Madaio, Shivani Kapania, Rida Qadri, Ding Wang, Andrew Zaldivar, Remi Denton, and Lauren Wilcox. 2024. Learning about Responsible AI On-The-Job: Learning Pathways, Orientations, and Aspirations. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency. 1544–1558.
  55. 55.Michael A Madaio, Jingya Chen, Hanna Wallach, and Jennifer Wortman Vaughan. 2024. Tinker, Tailor, Configure, Customize: The Articulation Work of Contextualizing an AI Fairness Checklist. Proceedings of the ACM on Human-Computer Interaction 8, CSCW1 (2024), 1–20.
  56. 56.Michael A Madaio, Luke Stark, Jennifer Wortman Vaughan, and Hanna Wallach. 2020. Co-designing checklists to understand organizational challenges and opportunities around fairness in AI. In Proceedings of the 2020 CHI conference on human factors in computing systems. 1–14.
  57. 57.Edward J. Markey. 2025. Artificial Intelligence Civil Rights Act of 2025. United States Senate. https://www.congress.gov/bill/119th-congress/senate-bill/3308 S. 3308, 119th Cong., 1st Sess..
  58. 58.New York City Council. 2021. New York City Local Law 144 of 2021 – Automated Employment Decision Tools (AEDT) Bias Audit Law. Municipal Legislation.
  59. 59.Ariadne A Nichol, Meghan Halley, Carole Federico, Mildred K Cho, and Pamela L Sankar. 2024. Moral engagement and disengagement in health care AI development. AJOB Empirical Bioethics 15, 4 (2024), 291–300.
  60. 60.Executive Office of the President. 2025. Ending Illegal Discrimination and Restoring Merit-Based Opportunity. Executive Order 14173. https://www.federalregister.gov/documents/2025/01/31/2025-02097/ending-illegal-discrimination-and-restoring-merit-based-opportunity 90 Fed. Reg. 8633 (Jan. 31, 2025).
  61. 61.Will Orr and Jenny L Davis. 2020. Attributions of ethical responsibility by Artificial Intelligence practitioners. Information, Communication & Society 23, 5 (2020), 719–735.
  62. 62.Samir Passi and Solon Barocas. 2019. Problem Formulation and Fairness. In Proceedings of the ACM Conference on Fairness, Accountability, and Transparency. 39–48.
  63. 63.Martha A. Poon. 2012. What Lenders See: A History of the Fair Isaac Scorecard. Ph. D. Dissertation. University of California. https://escholarship.org/uc/item/7n1369x2
  64. 64.Bogdana Rakova, Jingying Yang, Henriette Cramer, and Rumman Chowdhury. 2021. Where responsible AI meets reality: Practitioner perspectives on enablers for shifting organizational practices. Proceedings of the ACM on Human-Computer Interaction 5, CSCW1 (2021), 1–23.
  65. 65.Mark Ryan, Eleni Christodoulou, Josephina Antoniou, and Kalypso Iordanou. 2024. An AI ethics ‘David and Goliath’: value conflicts between large tech companies and their employees. AI & SOCIETY 39, 2 (2024), 557–572.
  66. 66.Seamus Ryan, Camille Nadal, and Gavin Doherty. 2023. Integrating fairness in the software design process: An interview study with hci and ml experts. IEEE Access 11 (2023), 29296–29313.
  67. 67.Malak Sadek and Celine Mougenot. 2024. Challenges in Value-Sensitive AI Design: Insights from AI Practitioner Interviews. International Journal of Human–Computer Interaction (2024), 1–18.
  68. 68.Nithya Sambasivan and Rajesh Veeraraghavan. 2022. The deskilling of domain expertise in AI development. In Proceedings of the 2022 CHI conference on human factors in computing systems. 1–14.
  69. 69.Jayshree Sarathy, Sophia Song, Audrey Haque, Tania Schlatter, and Salil Vadhan. 2023. Don’t look at the data! how differential privacy reconfigures the practices of data science. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–19.
  70. 70.Daniel S Schiff, Stephanie Kelley, and Javier Camacho Ibáñez. 2024. The emergence of artificial intelligence ethics auditing. Big Data & Society 11, 4 (2024), 20539517241299732.
  71. 71.ANDREW D SELBST and SOLON BAROCAS. 2023. UNFAIR ARTIFICIAL INTELLIGENCE: HOW FTC INTERVENTION CAN OVERCOME THE LIMITATIONS OF DISCRIMINATION LAW. University of Pennsylvania Law Review 171, 4 (2023).
  72. 72.Jessie J Smith, Anas Buhayh, Anushka Kathait, Pradeep Ragothaman, Nicholas Mattei, Robin Burke, and Amy Voida. 2023. The many faces of fairness: Exploring the institutional logics of multistakeholder microlending recommendation. In Proceedings of the 2023 ACM Conference on Fairness, Accountability, and Transparency. 1652–1663.
  73. 73.Jessie J Smith, Michael Madaio, Robin Burke, and Casey Fiesler. 2025. Pragmatic Fairness: Evaluating ML Fairness Within the Constraints of Industry. In Proceedings of the 2025 ACM Conference on Fairness, Accountability, and Transparency. 628–638.
  74. 74.Texas State Legislature. 2025. Texas Responsible Artificial Intelligence Governance Act, HB 149. State Law.
  75. 75.United States Congress. 1974. Equal Credit Opportunity Act, 15 U.S.C. §§ 1691–1691f. Federal Statute.
  76. 76.United States Congress. 1975. Home Mortgage Disclosure Act. Pub. L. No. 94-200, 89 Stat. 1124, codified at 12 U.S.C. §§ 2801–2810.
  77. 77.United States Congress. 1977. Community Reinvestment Act. Pub. L. No. 95-128, title VIII, 91 Stat. 1147, codified at 12 U.S.C. §§ 2901–2908.
  78. 78.Ari Ezra Waldman. 2021. Industry unbound: The inside story of privacy, data, and corporate power. Cambridge University Press.
  79. 79.John R. Walter. 1995. The Fair Lending Laws and Their Enforcement. Federal Reserve Bank of Richmond Economic Quarterly 81, 4 (1995), 61–77. https://www.richmondfed.org/~/media/richmondfedorg/publications/research/economic_quarterly/1995/fall/pdf/walter.pdf
  80. 80.Ding Wang, Shantanu Prabhat, and Nithya Sambasivan. 2022. Whose AI Dream? In search of the aspiration in data annotation.. In Proceedings of the 2022 CHI conference on human factors in computing systems. 1–16.
  81. 81.Qiaosi Wang, Michael Madaio, Shaun Kane, Shivani Kapania, Michael Terry, and Lauren Wilcox. 2023. Designing responsible ai: Adaptations of ux practice to meet responsible ai challenges. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems. 1–16.
  82. 82.David Gray Widder, Laura Dabbish, James D Herbsleb, and Nikolas Martelaro. 2024. Power and Play: Investigating"License to Critique"in Teams’ AI Ethics Discussions. Proceedings of the ACM on Human-Computer Interaction 8, CSCW2 (2024), 1–23.
  83. 83.David Gray Widder and Dawn Nafus. 2023. Dislocated accountabilities in the “AI supply chain”: Modularity and developers’ notions of responsibility. Big Data & Society 10, 1 (2023), 20539517231177620.
  84. 84.David Gray Widder, Dawn Nafus, Laura Dabbish, and James Herbsleb. 2022. Limits and possibilities for “Ethical AI” in open source: A study of deepfakes. In Proceedings of the 2022 ACM Conference on Fairness, Accountability, and Transparency. 2035–2046.
  85. 85.Lucas Wright, Roxana Mika Muenster, Briana Vecchione, Tianyao Qu, Pika (Senhuang) Cai, Alan Smith, Comm 2450 Student Investigators, Jacob Metcalf, and J. Nathan Matias. 2024. Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). Association for Computing Machinery, New York, NY, USA, 1701–1713. doi:10.1145/3630106.3658998
  86. 86.Lucas Wright, Roxana Mika Muenster, Briana Vecchione, Tianyao Qu, Pika (Senhuang) Cai, Alan Smith, Comm 2450 Student Investigators, Jacob Metcalf, and J. Nathan Matias. 2024. Null Compliance: NYC Local Law 144 and the challenges of algorithm accountability. In Proceedings of the 2024 ACM Conference on Fairness, Accountability, and Transparency (Rio de Janeiro, Brazil) (FAccT ’24). Association for Computing Machinery, New York, NY, USA, 1701–1713. doi:10.1145/3630106.3658998
  87. 87.Wenbin Zhang. 2024. AI fairness in practice: Paradigm, challenges, and prospects. Ai Magazine 45, 3 (2024), 386–395.

Citation

MLA
Black, E., et al. “The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice”. arXiv, 2026, http://arxiv.org/abs/2606.02957v3.
APA
Black, E., Bogen, M., Koepke, L., Barocas, S., Deng, W., & Hsu, M. (2026). The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice. arXiv. http://arxiv.org/abs/2606.02957v3
Chicago
Black, E., M. Bogen, L. Koepke, S. Barocas, W. Deng, and M. Hsu. 2026. “The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice”. arXiv. http://arxiv.org/abs/2606.02957v3.
Harvard
Black, E. et al. (2026) “The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice”, arXiv [Preprint]. Available at: http://arxiv.org/abs/2606.02957v3.
Vancouver
1. Black E, Bogen M, Koepke L, Barocas S, Deng W, Hsu M (2026) The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice. arXiv

BibTeX

@article{black2026the,
  title = {The Fair Lending Model: How the Longest-Running Algorithmic Fairness Programs Work in Practice},
  author = {Black, Emily and Bogen, Miranda and Koepke, Logan and Barocas, Solon and Deng, Wesley and Hsu, Mingwei},
  year = {2026},
  journal = {arXiv},
  url = {http://arxiv.org/abs/2606.02957v3},
  eprint = {2606.02957}
}
Metadata:arXiv

Access the Paper

This paper is available from its original source. Click below to access the PDF.

Open PDF
License: https://creativecommons.org/licenses/by/4.0/