By: Caroline Byrd*

I. Introduction.

Artificial intelligence (“AI”) has become a transformative force in the financial industry, particularly in mortgage lending, where algorithms are increasingly used to assess creditworthiness and manage risk.[1] AI in predictive mortgage lending refers to the use of advanced computational techniques, such as machine learning, natural language processing, and computer vision, to analyze vast datasets and forecast borrower behavior.[2] Proponents argue that AI can expand access to credit and reduce human bias by relying on objective data-driven processes, particularly in mortgage underwriting where such technologies are already in use.[3] However, AI lending systems often rely on historical data and non-traditional data points—such as geolocation, social media activity, and online shopping habits—which can inadvertently replicate and exacerbate the systemic inequities of traditional redlining.[4] In Baltimore, a city where the housing market has been deeply shaped by the legacy of redlining, these risks are especially pronounced.[5]

Redlining, the discriminatory practice of denying loans or insurance based on the racial composition of neighborhoods, was formally outlawed by the Fair Housing Act of 1968.[6] Yet, its impact persists, as historically marginalized communities continue to face limited access to financial resources and economic opportunity.[7] With the rise of AI, this exclusionary practice has taken on a new form—digital redlining—where algorithms unintentionally perpetuate the same patterns of discrimination embedded in historical data.[8] This is particularly evident in Baltimore, where decades of systemic inequities have left Black neighborhoods underinvested and economically disadvantaged.[9]  

This Comment explores how AI lending practices contribute to digital redlining, focusing on Baltimore as a case study.[10] Part II provides a historical background, examining how Baltimore became a model for redlining in the early 20th century, the long-term impact of housing segregation, and how the evolution of AI in mortgage lending has the potential to either mitigate or reinforce these discriminatory patterns.[11] Part III analyzes the key issues contributing to AI-driven redlining, including the reliance on alternative data, the risks of algorithmic bias, the lack of transparency in AI decision-making, and the challenges of regulating AI-driven lending practices under current legal frameworks.[12] Part IV then assesses the Maryland Online Data Privacy Act (“MODPA”) as a potential tool to address AI-driven discrimination.[13] However, this Comment argues that MODPA does not go far enough in regulating AI lending practices because it exempts credit reporting data, leaving a critical gap that allows financial institutions to continue using biased data in mortgage decisions.[14] Part IV proposes reforms to MODPA to ensure it effectively curbs modern redlining and provides meaningful consumer protections against AI-driven discrimination in lending.[15]

The central thesis of this Comment is that while AI holds the promise of innovation and efficiency in financial services, it also poses significant risks to equity and fairness, particularly in historically marginalized communities.[16] By leveraging MODPA and implementing additional safeguards, policymakers can create a regulatory framework that mitigates bias, promotes transparency, and ensures that AI serves as a tool for inclusion rather than exclusion.[17]

II. Historical Background.

A. The Birthplace of Redlining: Baltimore’s Role in Segregation Practices.

On December 19, 1910, the Mayor of Baltimore, J. Barry Mahool, passed an ordinance compelling the separation of races by residential zones.[18] As the first of its kind in the United States, it marked the beginning of a dark legacy for the city.[19] Citing the rationale of the ordinance, Mahool claimed that “Blacks should be quarantined in isolated slums in order to reduce the incidents of civil disturbance,” “prevent the spread of communicable diseases,” and “protect property values among the White majority.”[20] These ideas quickly permeated across the country inciting similar ordinances in large cities.[21]

Baltimore’s 1910 ordinance was issued in response to growing complaints from White neighbors after W. Ashbie Hawkins, a prominent African American lawyer, purchased a home on McCulloh Street, one of the city’s most desirable residential areas.[22] After Mahool’s 1910 ordinance, Hawkins became one of the leaders in advocating for racial equality.[23] He challenged racial zoning ordinances across the country, and in 1917 his efforts culminated in Buchanan v. Warley, where the Supreme Court struck down municipally mandated residential segregation.[24] In a majority opinion, the Supreme Court struck down racial zoning ordinances by holding them as an unconstitutional use of the state’s police power.[25] Fueled by this win, Hawkins continued to advocate for further advancements in racial equality across the country and at home in Baltimore.[26]

In the wake of Buchanan, segregationists had to resort to other methods of exclusion to ensure residential segregation.[27] A popular exclusionary method was restrictive covenants in deeds—language which prohibited the sale of the property to Black people.[28] Roland Park, a suburb outside of Baltimore, set forth yet another segregationist legacy by becoming one of the first neighborhoods in the U.S. to utilize racial restrictive covenants.[29] Deed restrictions became a way for White Baltimoreans to continue enforcing the racial zones established by Mahool’s 1910 ordinance.[30] The Federal Housing Administration (“FHA”) encouraged the use of racial restrictive covenants in its 1939 Underwriting Manual declaring them necessary “for a neighborhood to maintain any stability.”[31]

Racially restrictive covenants faced backlash from fair housing activists and as legal challenges multiplied across the country, the Supreme Court was pressured to address the issue in Shelley v. Kraemer.[32] The Court found that while the racial covenants were constitutional as they stand, judicial enforcement of the covenants violated the Equal Protection Clause of the Fourteenth Amendment due to the state action doctrine.[33]

The Supreme Court’s rulings in Buchanan and Kramer ousted two key methods of exclusionary zoning and forced wealthy White landowners to become less overt in their promotion of segregation.[34] Banks and realtors became leading actors for the promotion of segregation in the private sector by denying credit to Black homebuyers.[35] In the public sector, local governments, under the guise of urban renewal, constrained the expansion of Black neighborhoods encroaching on White institutions and used government money to build housing projects that confined displaced Black residents.[36]

The practice of denying credit to Black people further grew after the mapping of Baltimore by the Home Owner’s Loan Corporation (“HOLC”).[37] To recover from the economic impact of the Great Depression, the FHA developed a mortgage insurance program where a buyer could provide a ten percent downpayment on their home and the bank would issue a mortgage following a risk assessment.[38] This program extended only to the White homebuyer, further entrenching Black homebuyers in a cycle of poverty.[39] To determine creditworthiness, the FHA and HOLC created maps which designated certain residential areas with grades associated with their risk level.[40] This mapping method gave way to what we now refer to as “redlining”—the “D grade” or “hazardous” ratings given to predominately Black neighborhoods and colored red on the maps.[41]

Once a neighborhood received a “hazardous” rating, creditors routinely denied loans and Black buyers were forced to stay put or buy elsewhere.[42]  In effect, Black buyers who wanted to buy in “desirable” areas were denied access due to exorbitant pricing while simultaneously being denied loans in areas labeled “hazardous” by HOLC and therefore, forced to pay twenty percent more than their White counterparts.[43] Only two Black neighborhoods on HOLC’s map avoided the D rating.[44] The financial barrier pushed the Black population to the east and west sides of the city, while the White population occupied the desirable city center.[45] The resulting pattern forms what Lawrence Brown, professor at Morgan State University, coined as the “Black Butterfly” and the “White L”—whose shapes persist to this day.[46]

B. The Long-Term Effects of Housing Segregation Despite Federal Legal Improvements.

The Fair Housing Act of 1968 (the “Act”) was finally enacted in the fallout of Dr. Martin Luther King’s assassination and major civil rights movements.[47] The Act prohibited discrimination concerning the sale, rental, and financing of housing based on “race, color, religion, sex, national origin, familial status, and disability.”[48] In effect, racially motivated redlining in mortgage lending became illegal.[49] Another win for the Black community came after the enactment of the Equal Credit Opportunity Act in 1974 (“ECOA”), which expanded protections for minority groups by outlawing discrimination in mortgage transactions and requiring lenders to explain why applicants were denied.[50] However, these improvements in the law failed to end the cycle of poverty that was established by early racist housing practices and Baltimoreans have struggled to overcome these economic disparities to this day.[51]

In the seminal book American Apartheid, sociologistsDouglas Massey and Nancy Denton credit “opportunity hoarding” as a key reason why the cycle of segregation and poverty exists in cities like Baltimore.[52] Opportunity hoarding occurs when one group restricts access scarce resources through exclusionary tools.[53] This lack of resources is symptomatic of redlining and can include underfunded schools, food deserts, and lack of public transportation in addition to decreased credit lending.[54] According to the Urban Institute, between 2011 and 2016, predominately White neighborhoods in Baltimore received more than four times the investment of neighborhoods that are predominately Black.[55] By protecting the wealthiest population from economic competition, it becomes impossible for struggling neighborhoods to get the necessary traction they need for upward mobility, and the result is fewer and fewer middle class neighborhoods and, an ever widening class division.[56]

C. The Rise of AI in Mortgage Lending.

The rise of AI has had a great impact in a variety of industries such as healthcare, education, marketing, real estate, and banking.[57] This is due to AI’s ability to automate tasks and enhance decision-making which can have cost saving benefits.[58] Therefore, banks have implemented AI machine learning into their lending decisions in order to measure creditworthiness and manage risk.[59] Advocates of financial technology (“Fintech”) propose that computer algorithms running AI decision-making are less biased than humans.[60] However, this optimistic outlook has faced significant criticism, as many scholars argue that algorithmic processes can reinforce existing inequalities and evade fair lending laws due to their lack of transparency.[61] While decisions are made by computer algorithms, those algorithms are built by humans who have the ability to insert their own discriminatory biases, both conscious and unconscious.[62] These learning models are fed data that “can embed the same existing social disparities reflected in traditional scoring models.”[63] Machine learning models have the capability to pick out data points and carry out decision making in ways that go against lenders’ best practices.[64] Additionally, the algorithms operate within “black boxes,” offering little transparency to both lenders and borrowers who seek explanations for loan denials due to their adaptive and unpredictable nature.[65] Many have qualified algorithmic lending as “a new form of digital redlining” because the technology relies on pre-existing norms and “staple” decisions that tend to protect the status quo.[66] As law professor, Sheryll Cashin, simply stated: “technology exacerbates anti-Black bias.”[67]

III. Issue: AI in Mortgage Lending is Reinforcing Modern Redlining Through Bias and Reduced Transparency.

In Baltimore, where the effects of historical redlining are deeply entrenched, the implementation of AI in mortgage lending raises significant concerns.[68] For example, the use of alternative data can inadvertently disadvantage Black and low-income communities, as many of these data points may correlate with socio-economic factors tied to systemic discrimination.[69] Furthermore, this reliance on alternative data, such as a consumers social media network, can reinforce segregation patterns, effectively creating a digital version of redlining.[70] This Part examines how AI has become a central tool in the mortgage lending process, explores the inherent risks of bias and discrimination embedded within these technologies, and discusses the challenges to regulating these AI models.[71]

A. The Use of Alternative Data in AI Models.

Traditionally, creditworthiness was explicitly based on an individual’s credit score.[72] Now, in modern lending practices, loan officers can consider a wider pool of customer data known as “non-traditional data” when determining creditworthiness.[73] This non-traditional data consists of personal information, such as past rent payments, childcare payments, online shopping habits, social media usage, etc.[74] One of the main motivations for using more holistic alternative data, beyond traditional risk assessment, is to expand access to credit for borrowers who may not qualify under traditional credit scoring.[75] However, there are legitimate privacy concerns when using alternative data, such as website or social media activity, and additional ethical concerns raised due to the introduction of potential bias into the decision process.[76]

AI models take the use of non-traditional data a step further by analyzing vast datasets to identify patterns and predict borrower behavior.[77] Non-traditional data is collected by third-party “data brokers” that mine for personal data and sell it to credit bureaus.[78] Credit lenders then rely on machine learning algorithms to process data points such as zip codes, spending habits, or even where an applicant’s alma mater to determine creditworthiness.[79] Proponents argue that this allows for more comprehensive assessments and potentially reduces human bias.[80] The use of non-traditional data presents ethical concerns, as bias can be inadvertently introduced by programmers into AI algorithms at various stages of development.[81] Many data points used by these models serve as proxies for socio-economic or racial characteristics, which can inadvertently reinforce existing inequalities.[82] While the ECOA prohibits creditors from discriminating based on “race, color, religion, national origin, sex, marital status, age,” or receipt of public assistance, lenders now have a large universe of unprohibited proxy data that could lead to unintended discriminatory effects on minority groups.[83]

B. The Risks of Algorithmic Bias in AI Lending Models.

AI models, while powerful, are only as unbiased as the data they are trained on and the parameters programmed into them.[84] When AI systems are built on historical data or use non-traditional data as proxies, they can inadvertently replicate and even amplify existing disparities, creating significant risks of algorithmic bias in mortgage lending.[85] AI lending tools employ predictive algorithms which “are built by observing past patterns of behavior, and one of the enduring patterns in American economic life is the unequal distribution of opportunities along the lines of race, gender, and other personal characteristics.”[86] These algorithms in turn reflect decades of discriminatory practices, such as redlining.[87] For example, if an algorithm is trained using historical loan approval data that disproportionately excluded minority groups, it will learn to replicate those same exclusionary patterns.[88] This is particularly concerning in cities like Baltimore, where the legacy of redlining has left deep racial and economic divides.[89]

Therefore, AI systems built on biased data may further entrench these divides by denying loans or offering less favorable terms to residents of historically marginalized neighborhoods.[90]The impact of algorithmic bias extends beyond individual borrowers to entire communities, however, fair lending laws only protect against individual discrimination and not systemic discrimination.[91] In Baltimore, where minority neighborhoods are still struggling with underinvestment, biased AI models can perpetuate cycles of economic exclusion.[92] Residents of predominantly Black neighborhoods may face higher rates of loan denials or be offered less favorable terms, further limiting their ability to access credit and build wealth.[93] This not only reinforces systemic inequality[94] but also undermines the potential of AI to create a more equitable financial system.[95]

To address these risks and avoid perpetuating patterns of exclusion, such as the Black Butterfly in Baltimore, it is critical to recognize the inherent limitations of AI and to implement safeguards that mitigate algorithmic bias.[96] Without proper oversight, AI lending systems risk perpetuating the very inequalities they aim to resolve, particularly in vulnerable communities such as those in Baltimore.[97]

C. Lack of Transparency in Decision Making.

The “black box” issue commonly referenced regarding AI decision making refers to the lack of transparency that customers have.[98] The nature of the predictive algorithm is that it offers no insight into why a customer is approved or denied for a loan.[99] Lack of transparency is so prevalent in the Fintech industries that “almost seventy percent of companies [surveyed] could not explain how AI decisions predict certain outcomes.”[100] In response, the Consumer Financial Protection Bureau (“CFPB”) issued guidance Circular 2022-03, requiring all lenders to give specific reasoning for their decisions, in compliance with the ECOA, regardless of how “complex” and “opaque” their algorithms may be.[101] However, there has been no guidance as to how AI credit decision-making tools pass muster in order to comply with ECOA anti-discrimination standards.[102]

To establish a claim of lending discrimination under the Equal Credit Opportunity Act (“ECOA”), plaintiffs must demonstrate one of the following: (1) overt discrimination, where unlawful bias is evidenced through explicit and unambiguous statements; (2) disparate treatment, where lenders intentionally treat certain applicants less favorably than others based on a protected characteristic; or (3) disparate impact, where facially neutral lending practices result in a disproportionate and adverse effect on groups protected by the ECOA.[103] However, due to the lack of transparency in “black box” algorithms, it is nearly impossible to show that a plaintiff’s suffered injury was directly or proximately caused by an AI algorithm.[104] Therefore, complaining parties face issues collecting evidence sufficient to prove an ECOA violation or even negotiating more competitive loan terms.[105]

D. Regulatory Challenges in Addressing AI Bias.

The rise of AI in mortgage lending has outpaced the development of regulatory frameworks to address its unique challenges.[106] Existing laws, such as the FHA and the ECOA, prohibit discrimination in lending, but these frameworks were not designed with AI in mind.[107] As a result, they often fail to address the complexity of algorithmic bias and its impact on minority borrowers.[108] This regulatory gap creates significant challenges for ensuring fairness and equity in AI-driven lending practices.[109] Unlike traditional lending practices, which rely on static criteria, AI systems continuously adapt based on new data, making it challenging to establish clear regulatory standards.[110] Without a robust framework to oversee these systems, there is a risk that discriminatory practices will persist unchecked, particularly in cities like Baltimore, where historical inequities remain deeply embedded.[111]

AI faces increasing political barriers to regulation as well.[112] The Biden administration made headway with its AI Executive Order, which promoted “safe, secure, and trustworthy development” of AI and paved the way for states enacting their own AI executive orders and regulations.[113] However, following the transition in administration in 2025, President Trump rescinded Biden’s Executive Order on his first day in office and replaced it with his own directive, signaling a shift in federal AI policy priorities.[114] Additionally, Andrew Ferguson, the current FTC Chairman, as been vocal about “terminat[ing] all initiatives involving so called . . . AI ‘bias’”, “end[ing] politically motivated investigations,” and “not investigat[ing] conduct ‘under lawless disparate impact discrimination theories.’”[115] The changed political landscape will likely see an even larger push over the next three years to de-escalate policy initiatives for AI regulations over all industries, including Fintech.[116]

IV. Solution: MODPA Must Be Amended to Address AI Redlining in Mortgage Lending.

To address the challenges outlined above, Maryland state policymakers must adopt a proactive approach to regulating AI in lending.[117] This includes implementing transparency requirements, such as mandating that lenders provide explanations for AI-driven decisions and conducting regular audits of algorithms to identify and mitigate bias.[118] Furthermore, laws like MODPA could play a critical role in limiting the use of discriminatory data, ensuring that AI systems do not rely on variables that serve as proxies for race or socioeconomic status.[119] By addressing these regulatory gaps, policymakers can help ensure that AI lending models promote fairness and equity rather than exacerbating existing disparities.[120]

A. MODPA’s Exemption for Credit Reporting Data and its Implications.

The Governor of Maryland, Wes Moore, signed MODPA into law on May 9, 2024, to take effect on October 1, 2025.[121] The enactment of MODPA made Maryland the eighteenth state of twenty to adopt comprehensive data privacy legislation in the United States.[122] Overall, MODPA mirrors many of the other state data privacy acts with some notable differences, including unique data minimization requirements.[123] The data minimization provision in MODPA states that personal data collection must be limited “to what is reasonably necessary and proportionate to provide or maintain a product or service requested by the consumer to whom the data pertains.”[124] This means that MODPA primarily regulates data collection that extends beyond personal data deemed “reasonably necessary” to provide or maintain the requested product or service.[125] This safeguard to personal data has the potential to prevent the collection of alternative data points in lending that are unnecessary for determining creditworthiness (such as whether a consumer graduated college) and potentially curb digital redlining effects.[126]

However, a significant limitation of MODPA lies in its explicit exemption for credit reporting data.[127] This carve-out allows lenders to bypass MODPA’s stricter requirements when using data associated with credit reporting, such as credit scores and certain alternative data points tied to creditworthiness assessments.[128] While the intention behind this exemption may have been to align MODPA with existing federal frameworks, such as the Fair Credit Reporting Act (“FCRA”) and the Graham Leach Bliley Act (“GLBA”), it creates a regulatory gap that undermines MODPA’s broader goal of protecting consumer data.[129]

The CFPB warns that federal regulatory frameworks, such as FCRA and GLBA, do not properly protect consumers from modern data surveillance and that state laws like MODPA that carve out exceptions for financial institutions fail to address the gap in existing privacy laws.[130] The CFPB encourages state legislators to reconsider exempting financial institutions from their data privacy laws.[131] It is important to note, that the CFPB has advised state lawmakers that if they amend their data privacy laws to remove exceptions for financial institutions, they will not be preempted by FCRA and GLBA.[132] This is because both FCRA and GLBA expressly preserve from preemption state regulations consistent with, and that provide greater privacy protections than, their own requirements.[133] Therefore, as long as state privacy laws are not less protective than FCRA and GLBA, there is no risk of preemption.[134]

B. Evaluating MODPA’s Potential as a Tool Against AI-Driven Redlining.

One of MODPA’s most notable provisions is its data minimization requirement, which mandates that personal data collection be limited to what is “reasonably necessary and proportionate to provide or maintain a product or service requested by the consumer.”[135] In theory, this provision could  safeguard against AI lending models relying on excessive alternative data points—such as social media activity, educational background, and online purchases—that may reinforce existing biases.[136]

Digital civil rights groups, such as Access Now, have identified data minimization as a core principle of data protection.[137] The group opposes excessive corporate data collection practices that contribute to discriminatory outcomes.[138] Access Now recommends implementing robust data minimization requirements to mitigate discriminatory effects, while allowing a narrow exception for collecting data from protected groups solely for civil rights auditing aimed at identifying and eliminating harm.[139]

However, without specific guidance on AI lending, financial institutions may still justify collecting such data under the premise of improving risk assessment models.[140] Since AI models can detect correlations between non-traditional data and default risks, lenders may argue that these data points are “reasonably necessary” for evaluating creditworthiness, thereby skirting MODPA’s intended protections.[141] When making machine learning models, programmers decide “what variables to use, how to define categories or thresholds for sorting information, and which datasets to use to build algorithms.”[142] To address this issue, policymakers should consider amending MODPA to explicitly prohibit certain categories of alternative data from being collected, thereby pressuring programmers to recategorize which datasets to base their AI decision-making on.[143]

Another provision of MODPA prohibits the collection of consumer data for discriminatory purposes, which, on its face, could serve as an important tool in addressing AI-driven redlining.[144] If enforced correctly, this provision could prevent mortgage lenders from relying on data that disproportionately harms certain racial or economic groups.[145] However, the challenge lies in defining and enforcing discrimination in AI models.[146] Many AI lending systems do not explicitly factor in race or protected characteristics but rely on alternative data points that correlate with them, such as zip codes, commuting patterns, and social networks.[147] Because discrimination in AI lending often manifests as disparate impact rather than intentional bias, financial institutions may argue that their data collection practices are neutral even when they disproportionately exclude marginalized borrowers.[148] To make this provision effective, MODPA should be amended to require AI-specific audits and transparency reports.[149] Lenders should have to prove that their data collection and AI models do not result in discriminatory outcomes, particularly for historically marginalized communities like those in Baltimore.[150]

C. Strengthening Enforcement and Accountability.

Another weakness in MODPA is its lack of a private right of action, meaning that only the Maryland Attorney General (the “AG”) has the authority to enforce the law.[151] Without the ability for individuals or advocacy groups to bring lawsuits against non-compliant financial institutions, the effectiveness of MODPA’s protection is entirely dependent on the AG’s willingness to investigate and prosecute violations.[152]

This enforcement structure poses a substantial risk: if the AG does not prioritize AI-driven discrimination in lending, MODPA’s anti-discrimination provisions will remain largely symbolic.[153] Historically, fair lending violations have been difficult to detect and enforce due to the opacity of AI systems and the difficulty of proving discrimination.[154]

To strengthen enforcement, MODPA should be amended to include a private right of action, allowing consumers and civil rights organizations to challenge discriminatory lending practices in court.[155] This would increase accountability for financial institutions and provide borrowers with a legal avenue to contest unfair lending decisions.[156]

D. The Need for State Level Action in the Absence of Federal Oversight.

Trump’s second presidential term brings little hope for federal regulation of AI-driven lending discrimination.[157] In his last term as President, Trump successfully rolled back fair lending protections, including weakening the Disparate Impact Rule under the Fair Housing Act, making it more difficult for consumers to challenge discriminatory lending practices.[158] If this pattern continues, any hopes for federal oversight of AI lending models will be minimal, leaving states like Maryland to take the lead.[159] Maryland policymakers must recognize that state-level action is the primary means of protecting borrowers from AI-driven discrimination.[160] By strengthening MODPA to regulate AI lending practices, Maryland can fill the gap left by federal inaction and set a precedent for other states to follow.[161]

V. Conclusion.

AI has significantly transformed mortgage lending, offering enhanced efficiency and data-driven decision-making.[162] However, without proper oversight, AI-driven systems risk perpetuating discriminatory practices reminiscent of historical redlining, particularly in cities like Baltimore.[163] These systems may inadvertently rely on biased credit reporting data and alternative data points that serve as proxies for race and socioeconomic status, thereby reinforcing existing disparities.[164]

MODPA, enacted on May 9, 2024, aims to protect consumer data privacy.[165] While MODPA introduces robust data minimization requirements, it notably exempts credit reporting data from its scope.[166] This exemption permits lenders to continue utilizing potentially biased data without adhering to MODPA’s privacy safeguards, undermining the act’s intent to prevent discriminatory practices.[167] Moreover, MODPA does not provide a private right of action, granting exclusive enforcement authority to the AG.[168] This centralized enforcement mechanism may limit the act’s effectiveness, as it relies solely on the state’s initiative to address violations.[169]

Ultimately, addressing these issues is not just a matter of fairness but a necessary step toward ensuring economic equity in Baltimore and beyond.[170] By holding AI lending systems accountable and closing regulatory loopholes, Maryland has the opportunity to lead the nation in protecting its most vulnerable communities from the harms of modern redlining and advancing a more just financial future.[171]

---

* Caroline Byrd is a J.D. Candidate for May 2026 at the University of Baltimore School of Law and earned her B.A. in 2020 from James Madison University. She expresses her sincere gratitude to Professor Michele Gilman for her thoughtful guidance and mentorship throughout the development of this Comment. Professor Gilman’s work in poverty law and data privacy, particularly her focus on how emerging technologies impact marginalized communities, was instrumental in shaping the analysis of AI-driven redlining in mortgage lending. She also thanks the University of Baltimore Law Forum Editorial Board and Staff Editors for their time, care, and dedication in preparing this Comment for publication. Finally, she extends her appreciation to her family and friends for their constant encouragement and support throughout law school and the writing process.

[1]   Roman Bevz, The Role of AI and ML in Transforming Credit Risk Mgmt. in Banking, Avenga (Apr. 29, 2025), https://www.avenga.com/magazine/ai-for-credit-risk-management (on file with the University of Baltimore Law Forum). Mortgage lending refers to the process by which financial institutions extend credit to borrowers for the purchase or refinancing of real property, typically secured by the property itself. Mortgages: Types, How They Work, and Examples, Investopedia (Dec. 21, 2025), https://www.investopedia.com/terms/m/mortgage.asp (on file with the University of Baltimore Law Forum).

[2]   Kris van Beever, All Abord the AI Train: Prac. Roadmap for Lenders, Stratmor Group (Feb. 2025), https://www.stratmorgroup.com/all-aboard-the-ai-train-a-practical-roadmap-for-lenders (on file with the University of Baltimore Law Forum).

[3]   Michele Gilman, The Impact of Proptech and the Datafication of Real Estate on the Human Right to Housing, 9 Geo. L. Tech. Rev. 444, 469 (citing William Magnuson, Artificial Financial Intelligence, 10 Harv. Bus. L. Rev. 337, 349 (2020)).

[4]   Maria Sarah, Using Alt. Data and Artificial Intelligence to Expand Fin. Inclusion: Evidence Based Insights, J-PAL (Mar. 21, 2024), https://www.povertyactionlab.org/blog/3-21-24/using-alternative-data-and-artificial-intelligence-expand-financial-inclusion-evidence#:~:text=In%20contrast%20to%20traditional%20credit,data%E2%80%9D%E2%80%94to%20make%20predictions (on file with the University of Baltimore Law Forum).

[5]   Dario Toval, Was Baltimore the Proving Ground for Redlining?, Reading Partners (June 12, 2024), https://readingpartners.org/blog/was-baltimore-the-proving-ground-for-redlining/#:~:text=People%20in%20historically%20redlined%20areas,communities%20never%20having%20been%20disrupted (on file with the University of Baltimore Law Forum).

[6]   Id.

[7]   Id.

[8]   Jason Jia-Xi Wu, Algorithmic Fairness in Consumer Credit Underwriting: Towards a Harm-Based Framework for AI Fair Lending, 21 Berkeley Bus. L.J. 65, 90 (2024).

[9]   Toval, supra note 5.

[10] See infra Parts II-III.

[11] See infra Part II.

[12] See infra Part III.

[13] See infra Part IV.

[14] See infra Part IV.

[15] See infra Part IV.

[16] See infra Part III; Gilman, supra note 3, at 3-4.

[17] See infra Part IV.

[18] Baltimore Tries Drastic Plan of Race Segregation, N.Y. Times (Dec. 25, 1910), https://timesmachine.nytimes.com/timesmachine/1910/12/25/105900067.html?pageNumber=34 (on file with the University of Baltimore Law Forum).

[19] Id.

[20] Sheryll Cashin, White Space, Black Hood: Opportunity Hoarding and Segregation in the Age of Inequality 10 (2021).

[21] Id.

[22] Id.

[23] See generally Domonique Flowers, Passing the Mantle: The Transformative Initiatives of Harry Commings, Ashbie Hawkins, and the Next Generation of African American Lawyers in Maryland, to Utilize the Legal System and Advance the Racial Progress of African American Citizens, 54.1 U. Balt. L. F. 22 (2023) (describing Hawkins’ efforts in advocating for racial equality).

[24] Cashin, supra note 20, at 12.

[25] See generally Buchanan v. Warley, 245 U.S. 60, 82 (1917) (holding that a racial zoning ordinance was an unconstitutional exercise of the state’s police power in violation of the Fourteenth Amendment).

[26] Cashin, supra note 20, at 12.

[27] Id. at 15.

[28] Id.

[29] Id. at 13.

[30] Id. at 15.

[31] Sarah Jacobson, A Brief Hist. of Hous. Segregation in Balt., Intl. Mapping (Oct. 29, 2021), https://internationalmapping.com/blog/a-brief-history-of-housing-segregation-in-baltimore/ (on file with the University of Baltimore Law Forum).

[32] Douglas S. Massey & Nancy A. Denton, American Apartheid: Segregation and the Making of the Underclass 188 (8th ed. 1998).

[33] See generally Shelly v. Kraemer, 334 U.S. 1, 20 (1948) (explaining that although private parties may enter such agreements, the Constitution prohibits the state from giving them legal effect through judicial enforcement).

[34] Cashin, supra note 20, at 13.

[35] Id.at 14.

[36] Massey & Denton, supra note 32, at 188-89.

[37] Cashin, supra note 20, at 15.

[38] Jacobson, supra note 31.

[39] Id.

[40] Id.

[41] Cashin, supra note 20, at 14-15.

[42] A Brief Hist. of Redlining, NYC.Gov (Jan. 2, 2021), https://a816-dohbesp.nyc.gov/IndicatorPublic/data-stories/redlining (on file with the University of Baltimore Law Forum).

[43] Cashin, supra note 20, at 14-15.

[44] Id. at at 14.

[45] Id. at 15.

[46] Marceline White, Baltimore: the Black Butterfly, Natl. Comty. Reinvestment Coalition (Oct. 8, 2020), https://ncrc.org/the-black-butterfly (on file with the University of Baltimore Law Forum)

[47] Hist. of Fair Hous., U.S. Dept. OF Hous. and Urban Dev., https://www.hud.gov/program_offices/fair_housing_equal_opp/aboutfheo/history  (on file with the University of Baltimore Law Forum) (last visited Nov. 18, 2024).

[48] 42 U.S.C. § 3604 (1988).

[49] Jacobson, supra note 31.

[50] 15 U.S.C. § 1691, 1691(d) (2010).

[51] Jacobson, supra note 31.

[52] Massey & Denton, supra note 32.

[53] Id.

[54] White, supra note 46. See generally Cashin, supra note 20, at 21–37 (2021) (chronicling the transit activism in Baltimore during the Hogan Administration).

[55] Lionel Foster, The Black Butterfly, Racial Segregation and Inv. Patterns in Baltimore, Urban Inst. (Feb. 5, 2019), https://apps.urban.org/features/baltimore-investment-flows (on file with the University of Baltimore Law Forum).

[56] Cashin, supra note 20, at 119.

[57] Akash Takyar, AI Use Cases & Applications Across Major Indus., LeewayHertz, https://www.leewayhertz.com/ai-use-cases-and-applications (on file with the University of Baltimore Law Forum) (last visited July 13, 2024).

[58] Id.

[59] Id.

[60] Gilman, supra note 3, at 469 (first citing Matthew Adam Bruckner, The Promise and Perils of Algorithmic Lenders’ Use of Big Data, 93 Chi.-Kent L. Rev. 3, 11-13 (2018); and then citing Vincent Di Lorenzo, Fintech Lending: A Study of Expectations Versus Market Outcomes, 38 Rev. Banking & Fin. L. 725, 728 (2019)).

[61] Wu, supra note 8, at 69-70.

[62] Gilman, supra note 3, at 469 (citing Sian Townson, AI Can Make Bank Loans More Fair, Harv. Bus. Rev. (Nov. 6, 2020), https://hbr.org/2020/11/ai-can-make-bank-loans-more-fair (on file with the University of Baltimore Law Forum)); see also Andreas Fuster et al., Predictably Unequal? The Effects of Machine Learning on Credit Markets, 67 J. Fin. 5 (2022), https://doi.org/10.1111/jofi.13090 (on file with the University of Baltimore Law Forum) (demonstrating that machine learning in credit markets can produce disparate outcomes by incorporating patterns from historically biased data).

[63] Gilman, supra note 3, at 469.

[64] Wu, supra note 8, at 91.

[65] Id. at 70-71; Gilman, supra note 3, at 470. A “black box” model is one in which the internal logic of the algorithm is not readily interpretable, limiting transparency into how outputs are generated from given inputs. Gilman, supra note 3, at 470.

[66] Wu, supra note 8, at 90-91.

[67] Cashin, supra note 20, at 189.

[68] See, e.g., Jacobson, supra note 31; see supra Section III.B.

[69] Lorena Rodriguez, All Data is not Credit Data: Closing the Gap Between the Fair Housing Act and Algorithmic Decisionmaking in the Lending Industry, 120 Colum. l. Rev. 1843, 1857 (2020).

[70] See Robinson Meyer, Could a Bank Deny Your Loan Based on Your Facebook Friends?, The Atlantic (Sep. 25, 2015), https://www.theatlantic.com/technology/archive/2015/09/facebooks-newpatent-and-digital-redlining/407287/ (on file with the University of Baltimore Law Forum).

[71] See infra Part III.

[72] Tom Sullivan, 6 Types of Alt. Credit Data for Better Loan Decisions, Plaid (Feb. 25, 2025), https://plaid.com/resources/lending/alternative-credit-data (on file with the University of Baltimore Law Forum).

[73] Terri Bradford, Give me Some Credit: Using Alt. Data to Expand Credit Access, Fed. Rsrv. Bank of Kan. City (June 28, 2023), https://www.kansascityfed.org/research/payments-system-research-briefings/give-me-some-credit-using-alternative-data-to-expand-credit-access (on file with the University of Baltimore Law Forum).

[74] Id.

[75] Sullivan, supra note 72.

[76] Id.

[77] Wu, supra note 8, at 78 (noting that alternative data, including online activity and behavioral data, may involve the collection of sensitive information without clear consumer awareness or consent, raising concerns about surveillance and data misuse).

[78] Id.

[79] Rodriguez, supra note 69, at 1856, 1858.

[80] Gilman, supra note 3, at 469.

[81] Rodriguez, supra note 69, at 1856.

[82] Id. at 1857.

[83] The Equal Credit Opportunity Act, U.S. Dep’t of Just., https://www.justice.gov/crt/equal-credit-opportunity-act-3 (on file with the University of Baltimore Law Forum) (last visited Jan. 2, 2025); Rodriguez, supra note 69, at 1860-61. 

[84] Gilman, supra note 3, at 469.

[85] Rodriguez, supra note 69, at 1857.

[86] Gilman, supra note 3, at 474 (quoting Pauline T. Kim, Manipulating Opportunity, 106 Va. L. Rev. 867, 869–70 (2020)).

[87] Wu, supra note 8, at 90.

[88] Gilman, supra note 3, at 469. 

[89] Jacobson, supra note 31; Gilman, supra note 3, at 468.

[90] Wu, supra note 8, at 91.

[91] Id. at 92 (noting that federal fair lending laws, such as the Equal Credit Opportunity Act and the Fair Housing Act, are primarily designed to address discrimination against individual applicants or protected classes, and are less equipped to address systemic or structural discrimination that arises from facially neutral policies or algorithms).

[92] Jacobson, supra note 31.

[93] Sienna Duran-Kneip, Black, White, and Redlining All Over: How Maryland’s Appraisal Gap From Historic Redlining Financial Assistance Program Aims to Ameliorate the Lasting Effects of Discriminatory Property Practices, U. Balt. L.F. (Apr. 26, 2024), https://ublawforum.com/2024/04/26/appraisal-gap/ (on file with the University of Baltimore Law Forum).

[94] Gilman, supra note 3, at 472.

[95] Regina Curry & Deep Ratna Srivastav, Dimensions & Insights: AI’s Transformative Impact—Broadening Financial Inclusion, Franklin Templeton (July 23, 2024), https://www.franklintempleton.com/articles/2024/strategist-views/dimensions-insights-ais-transformative-impact-broadening-financial-inclusion (on file with the University of Baltimore Law Forum).

[96] Rodriguez, supra note 69, at 1882.

[97] Omogbeme Angela & Oyindamola Modupe Odewuyi, AI-Powered DEI Metrics in Financial Institutions: Driving Inclusive Growth, 5 Intl. J. of Rsch. Publ’n & Revs. 3089, 3094 (2024). 

[98] Lou Blouin, AI’s Mysterious ‘Black Box’ Problem, Explained, U. of Mich.-Dearborn News (Mar. 6, 2023), https://umdearborn.edu/news/ais-mysterious-black-box-problem-explained (on file with the University of Baltimore Law Forum).

[99] Gilman, supra note 3, at 470. 

[100] Id. (citing Jonathan Greig, Report Finds Startling Disinterest in Ethical, Responsible Use of AI Among Business Leaders, ZDNET (May 25, 2021, at 05:00 PT), https://www.zdnet.com/article/fico-report-finds-startling-disinterest-in-ethical-responsible-use-of-ai-among-business-leaders/ (on file with the University of Baltimore Law Forum)).

[101] Consumer Financial Protection Circular 2022-03, CFPB (May 26, 2022), https://www.consumerfinance.gov/compliance/circulars/circular-2022-03-adverse-action-notification-requirements-in-connection-with-credit-decisions-based-on-complex-algorithms/ (on file with the University of Baltimore Law Forum).

[102] Wu, supra note 8 at, 125-26.

[103] Winnie Taylor, Proving Discrimination and Monitoring Fair Lending Compliance: The Missing Data Problem in Nonmortgage Credit, 31 Rev. of Banking & Fin. L. 199, 208-11 (2011-2012).

[104] Rodriguez, supra note 69, at 1872.

[105] Wu, supra note 8, at 133.

[106] Michael Neal et al., AI Could Alter Mortgage Lending, but Government Leadership Is Needed, Urban Inst. (Nov. 6, 2023), https://www.urban.org/urban-wire/ai-could-alter-mortgage-lending-government-leadership-needed (on file with the University of Baltimore Law Forum).

[107] Wu, supra note 8, at 69.

[108] Id.

[109] Id. at 125.

[110] Id. at 124.

[111] Id. at 87.

[112] Marc S. Martin & Sydney Veatch, What to Expect from the New Administration on AI Policy, Perkins Coie (Dec. 6, 2024), https://perkinscoie.com/insights/update/what-expect-trump-administration-ai-policy (on file with the University of Baltimore Law Forum).

[113] FACT SHEET: President Biden Issues Executive Order on Safe, Secure, and Trustworthy Artificial Intelligence, The White House (Oct. 30, 2023), https://bidenwhitehouse.archives.gov/briefing-room/statements-releases/2023/10/30/fact-sheet-president-biden-issues-executive-order-on-safe-secure-and-trustworthy-artificial-intelligence/ (on file with the University of Baltimore Law Forum).

[114] David Shepardson, Trump revokes Biden Executive Order on Addressing AI risks, Reuters (Jan. 21, 2025), https://www.reuters.com/technology/artificial-intelligence/trump-revokes-biden-executive-order-addressing-ai-risks-2025-01-21 (on file with the University of Baltimore Law Forum) (reporting that President Trump revoked Biden’s AI Executive Order addressing safety and bias concerns and issued a new order prioritizing deregulation and limiting federal oversight of artificial intelligence).

[115] Id. (quoting Ferguson’s leaked FTC agenda on social media platform X); Shylah R. Alfonso et al., Insight into the Upcoming Trump Administration’s Antitrust Policy, Perkins Coie (Jan. 10, 2025), https://perkinscoie.com/insights/update/insight-upcoming-trump-administrations-antitrust-policy (on file with the University of Baltimore Law Forum)).

[116] Martin & Veatch, supra note 112.

[117] See, e.g., Consumer Financial Protection Circular 2022-03, CFPB (May 26, 2022), https://www.consumerfinance.gov/compliance/circulars/circular-2022-03-adverse-action-notification-requirements-in-connection-with-credit-decisions-based-on-complex-algorithms (on file with the University of Baltimore Law Forum) (emphasizing the need for oversight of algorithmic decision-making in lending); Cary Coglianese & Alicia Lai, Algorithm vs. Algorithm, 71 Duke L.J. 1281, 1333-39 (2021) (highlighting the need for regulatory approaches to manage risks posed by algorithmic systems); White House Off. of Sci. & Tech. Pol’y, Blueprint for an AI Bill of Rights: Making Automated Systems Work for the American People 5 (2022), https://marketingstorageragrs.blob.core.windows.net/webfiles/Blueprint-for-an-AI-Bill-of-Rights.pdf (on file with the University of Baltimore Law Forum) (calling for safeguards against algorithmic discrimination).

[118] See infra Part IV.

[119] See infra Part IV.

[120] See infra Part IV.

[121] F. Paul Pittman & Abdul M. Hafiz, Md. Enacts Comprehensive Data Privacy Law, White & Case (May 14, 2024), https://www.whitecase.com/insight-alert/maryland-enacts-comprehensive-data-privacy-law (on file with the University of Baltimore Law Forum).

[122] Which States Have Consumer Data Privacy Laws?, Bloomberg L. (Apr. 7, 2025), https://pro.bloomberglaw.com/insights/privacy/state-privacy-legislation-tracker/#map-of-state-privacy-laws (on file with the University of Baltimore Law Forum).

[123] Kate Lucente et. al., US: Maryland Online Data Privacy Act Summary and Comparative Analysis, DLA Piper (July 2, 2024), https://www.dlapiper.com/en-us/insights/publications/2024/07/us-maryland-online-data-privacy-act-summary-and-comparative-analysis (on file with the University of Baltimore Law Forum).

[124] H.B. 567, 2024 Gen. Assemb., 446th Sess. (Md. 2024).

[125] David Stauss, Maryland Legislature Passes Consumer Data Privacy Bill, Husch Blackwell (Apr. 8, 2024), https://www.bytebacklaw.com/2024/04/maryland-legislature-passes-consumer-data-privacy-bill (on file with the University of Baltimore Law Forum).

[126] See supra Section III.A.

[127] H.B. 567 § 14-4603(B), 2024 Gen. Assemb., 446th Sess. (Md. 2024).

[128] See Report Details Carveouts for Financial Institutions in State Data Privacy Laws, CFPB (Nov. 12, 2024), https://www.consumerfinance.gov/about-us/newsroom/cfpb-report-details-carveouts-for-financial-institutions-in-state-data-privacy-laws/#:~:text=The%20current%20federal%20framework%20for,collection%20of%20especially%20sensitive%20data (on file with the University of Baltimore Law Forum).

[129] Id.

[130] Id.

[131] Id.

[132] Kevin M. Toomey et al., CFPB Report Highlights Gaps in Privacy Protections for Financial Services Consumers; Suggests State Action, Arnold & Porter (Dec. 11, 2024), https://www.arnoldporter.com/en/perspectives/advisories/2024/12/cfpb-report-highlights-gaps-in-privacy-protections#:~:text=In%20the%20CFPB’s%20view%2C%20neither,than%20the%20federal%20laws’%20requirements (on file with the University of Baltimore Law Forum).

[133] Id.

[134] Id.

[135] Stauss, supra note 125.

[136] See supra Part III (referencing the overuse of alternative data in lending decisions).

[137] Eric Null et al., Data Minimization Report: Key to Protecting Privacy and Reducing Harm, Access Now 4 (May 2021), https://www.accessnow.org/wp-content/uploads/2021/05/Data-Minimization-Report.pdf (on file with the University of Baltimore Law Forum).

[138] Id. at 7-8.

[139] Id.

[140] See supra Section III.B; Ritesh Shetty, The Role of AI in Transforming Credit Risk Assessment Process, Arya.ai (May 30, 2024), https://arya.ai/blog/ai-in-credit-risk-assessment (on file with the University of Baltimore Law Forum).

[141] Shetty, supra note 140.

[142] A.R. Lange & Natasha Duarte, Understanding Bias in Algorithmic Design, Medium (Sept. 6, 2017), https://medium.com/impact-engineered/understanding-bias-in-algorithmicdesign-db9847103b6e (on file with the University of Baltimore Law Forum).

[143] See supra Part III and Section IV.A (discussing the benefits of amending MODPA).

[144] See generally H.B. 567 § 14-4607, 2024 Gen. Assemb., 446th Sess. (Md. 2024) (prohibiting the collection of consumer data for discriminatory purposes).

[145] Id.

[146] See supra Section III.B (explaining the transparency challenges between the back-end algorithm and the loan decision).

[147] See supra Section III.A (explaining the migration to alternative data collection from traditional data collection).

[148] Wu, supra note 8, at 92 (explaining the need to move past typical disparate impact liability when examining algorithmic discrimination issues).

[149] See supra Section III.C.

[150] See supra Section III.C (showing the need for more transparency in algorithmic decision-making).

[151] Lucente et al., supra note 123.

[152] Id.

[153] Id.

[154] See supra Section III.C (discussing the “black box” issue).

[155] Becky Chao et al., Enforcing a New Privacy Law: A Private Right of Action is Key to Ensuring that Customers Have their Own Avenue for Redress, New Am. 2, 16 (Nov. 20, 2019), https://www.newamerica.org/oti/reports/enforcing-new-privacy-law/a-private-right-of-action-is-key-to-ensuring-that-consumers-have-their-own-avenue-for-redress/#:~:text=A%20private%20right%20of%20action%20is%20critical%20for%20aggrieved%20individuals,relying%20on%20the%20federal%20enforcer (on file with the University of Baltimore Law Forum).

[156] Id.

[157] Martin & Veatch, supra note 112.

[158] Izzy Woodruff & Phoebe Plagens, National Fair Housing Alliance Challenges Harmful Trump Administration Reversal of Fair Housing Rule, NFHA (Oct. 22, 2020), https://nationalfairhousing.org/national-fair-housing-alliance-challenges-harmful-trump-administration-reversal-of-fair-housing-rule (on file with the University of Baltimore Law Forum).

[159] Danielle Ochs et al., Trump Administration Unveils New AI Policy, Reverses Biden’s Regulatory Framework, Ogletree Deakins (Feb. 7, 2025), https://ogletree.com/insights-resources/blog-posts/trump-administration-unveils-new-ai-policy-reverses-bidens-regulatory-framework (on file with the University of Baltimore Law Forum).

[160] See generally Seth Frothman & Brian Shearer, Strengthening State-Level Consumer Protections, CFPB (Jan. 14, 2025), https://www.consumerfinance.gov/about-us/blog/strengthening-state-level-consumer-protections (on file with the University of Baltimore Law Forum) (highlighting the role of states in advancing consumer protection efforts, particularly in areas where federal regulation may be insufficient).

[161] Toomey, supra note 132.

[162] See supra Section II.C (laying the background on technological advancements in the mortgage industry).

[163] See supra Section III.B.

[164] See supra Section III.B.

[165] See supra Section IV.A.

[166] See supra Section IV.A.

[167] See supra Section IV.A.

[168] See supra Section IV.C.

[169] See supra Section IV.C.

[170] See supra Sections II.C and III.B.

[171] See supra Section IV.D.