Your bank does not want you to understand how it decides.
This is not a conspiracy theory. It is a description of the business model. The traditional lending system—mortgages, credit cards, auto loans, personal loans—operates on a simple principle: the lender knows the rules, and you do not. The credit score is proprietary. The underwriting model is a trade secret. The reasons for denial are communicated in vague language designed to be technically compliant with disclosure laws while remaining functionally incomprehensible. You are scored, ranked, and priced by a system you cannot inspect, cannot question, and cannot audit.
Decentralized finance does not solve every problem. It introduces new ones. But it does one thing that the traditional system cannot do and will never voluntarily do: it makes the mechanism visible. Every lending parameter, every interest rate model, every collateral requirement, every liquidation threshold is written in open-source code on a public blockchain, where anyone can read it, verify it, and audit it. The algorithm is not a black box. It is a public document.
That is not a minor difference. It is the difference between a system that can be held accountable and a system that cannot.
The Traditional System: Opacity as Architecture
Start with the score. The FICO score is the most consequential number in most Americans’ financial lives. It determines whether you get a mortgage, what interest rate you pay, whether you can rent an apartment, and in some states, whether you can get a job. The precise formula is proprietary. The weights assigned to each factor—payment history, amounts owed, length of credit history, new credit, credit mix—are published only in broad outline. The actual model is a trade secret.
What we do know about how the system performs is damning. Black mortgage applicants are denied at 27.11 percent, compared to substantially lower rates for white and Asian borrowers (National Fair Housing Alliance, 2025). The Philadelphia Federal Reserve found that Black and Hispanic borrowers are “substantially more likely to have their loan application rejected” and “substantially more likely to default on their mortgages than White Americans” (Elzayn et al., 2025). A Financial Times analysis found that Black Americans are almost twice as likely to be turned away for mortgages as white applicants with the same income (Operation HOPE, 2025). In every Deep South state, Black denial rates range from 28 to 36 percent, versus 10 to 20 percent for white applicants (Kindred Futures, 2025).
The standard defense is that these disparities reflect underlying risk factors. But the research does not support that. The HUD-Treasury National Predatory Lending Task Force documented in 2000 that subprime loans were disproportionately concentrated in Black neighborhoods compared to white neighborhoods, even after controlling for income (Federal Reserve, 2022). A study found that borrowers of color were more than 30 percent more likely to receive a higher-rate loan than white borrowers even after accounting for differences in creditworthiness (Federal Reserve, 2022). The disparities are not explained by risk. They are explained by race.
And then there is the algorithmic layer. Machine learning models in credit decisioning “disproportionately lessen the access to financing for Black and Hispanic borrowers” (Fuster et al., 2021, as cited in Management System Engineering, 2025). The models are trained on historical data that encodes historical discrimination. A lending AI “could unintentionally eliminate applicants who belong to specific postal areas or backgrounds even though each applicant poses no risk of default because these areas show higher default rates in historical data, which amounts to digital redlining” (Cambridge, 2025).
The opacity is not incidental. It is what makes the discrimination possible. If you cannot see the model, you cannot prove the bias. If you cannot prove the bias, you cannot challenge it. The black box is not a technical limitation. It is a legal shield.
The European Variation: Fragmentation as Opacity. The European lending market adds a layer of opacity that is structurally distinct from the American system. In the United States, the problem is a single proprietary model applied across a unified market. In Europe, the problem is a patchwork of national credit bureaus, each with its own scoring model, its own data format, and its own scale—and none of them interoperable (Kreditai, 2026). Every country runs its own system, and the scores do not travel. A Romanian software engineer with a decade of clean credit history, a paid-off mortgage, and a permanent contract in Amsterdam was rejected for a €5,000 personal loan because the Dutch credit register had no record of him (Kreditai, 2026). Roughly 20 million EU residents live in a member state other than their country of birth, and another 1.4 million move within the EU for work every year; the total number of people with financial ties to more than one European country sits at about 45 million—and most of them will hit this wall (Kreditai, 2026). Being foreign makes your money more expensive, even if you are a better borrower than the local average, because the lender cannot see your history and prices the uncertainty into the rate (Kreditai, 2026).
The opacity is not merely technical. It is legal. The German credit agency Schufa, which dominates the German market, keeps parts of its calculation basis confidential—and a petition to the European Parliament argued that this violates democratic principles, that low-income households are discriminated against because citizens living in structurally weak regions are rated lower, and that a large proportion of citizens with migration backgrounds are disadvantaged in loan decisions (European Parliament, 2024). The petition was declared admissible. The Commission’s response was that the Consumer Credit Directive does not regulate credit scoring as such (European Parliament, 2024). The directive that will apply from November 2026, Directive 2023/2225, requires creditworthiness assessments to be conducted “in the interest of the consumer” and to consider “relevant and accurate information” about income, expenses, and financial circumstances—but it still does not require lenders to disclose the scoring model itself (European Parliament, 2024).
The courts are beginning to intervene. In February 2025, the Court of Justice of the European Union ruled in the Dun & Bradstreet Austria case that a data subject is entitled to an explanation of how an automated credit decision was taken, and that the explanation “must enable the data subject to understand and challenge the automated decision” (CJEU, 2025). The Court held that the controller must describe the procedure and principles actually applied “in such a way that the data subject can understand which of his or her personal data have been used, and how they have been used,” and that the mere communication of an algorithm does not constitute a sufficiently concise and intelligible explanation (CJEU, 2025). Where the controller claims trade secrets, it must provide the allegedly protected information to the supervisory authority or court, which will balance the rights and interests at issue (CJEU, 2025). The ruling is a step forward. But it applies to the explanation of an individual decision, not to the structure of the model that produced it. The model itself remains opaque.
The European fragmentation has a second consequence. Because the scoring systems are national, the bias they encode is also national. A model trained on German data encodes German historical patterns—including the historical exclusion of migrant communities and the structural underdevelopment of certain regions. When that model is applied to a Turkish-German applicant or a resident of a deindustrialized eastern district, it reproduces the patterns of the past as if they were predictions of the future (European Parliament, 2024). The same logic applies across every member state. The fragmentation does not protect against bias. It localizes it, embedding it in the infrastructure of each national market and making cross-border comparison—and therefore cross-border accountability—impossible.
The European AI Act lists credit scoring as a high-risk AI system, subject to conformity assessment requirements that enforce from August 2026 (Zenodo, n.d.). But the Act regulates risk management, not disclosure. It requires that high-risk systems be designed to allow human oversight and be accompanied by instructions for use. It does not require that the scoring model be published, that the weights be disclosed, or that the applicant be told which variables drove the decision. The model remains a black box. The box is now certified. That is not the same thing as transparent. The European borrower, like the American borrower, is scored by a mechanism they cannot read. The difference is that in Europe, the mechanism is also different in every country they move to—and the score they built over a decade in one member state is worth nothing in the next.
DeFi: The Mechanism Made Visible
Decentralized lending protocols like Aave and Compound operate on a fundamentally different principle. The lending logic is not a trade secret. It is open-source code deployed on a public blockchain. Every parameter is visible. Every transaction is recorded. Every liquidation is auditable.
The core mechanism is overcollateralization. A borrower deposits an asset as collateral and borrows against it. The loan-to-value ratio—the maximum percentage of the collateral’s value that can be borrowed—is set by governance and written into the smart contract. On Aave V3 and Compound, the minimum collateral ratio is typically 120 percent or higher (econstor, n.d.). The liquidation threshold, the point at which the position becomes eligible for liquidation, is also a public parameter. Compound III’s documentation describes its liquidation collateral factors as “separate and higher than” its borrow collateral factors, which leaves a price buffer on every new position (Proof of Talk, 2026). Morpho Blue works with a single number per market, the liquidation loan-to-value (Proof of Talk, 2026).
The interest rate model is equally transparent. Aave V3 uses a two-slope model around an optimal utilization point: below the kink, borrow rates rise along the first slope; above it, they rise faster along the second (Proof of Talk, 2026). Compound III uses the same shape (Proof of Talk, 2026). The rate does not depend on who the borrower is. It depends on how much of the pool is lent out. There is no credit score. There is no underwriting committee. There is no proprietary model. There is a formula, and the formula is public.
This is not a minor advantage. It is a categorical difference. In the traditional system, the borrower is scored by an algorithm they cannot see, using data they cannot verify, to produce a decision they cannot challenge. In DeFi, the borrower interacts with a mechanism they can read, verify, and audit before they ever deposit a single token. The algorithm does not trust you. It does not need to. It does not know who you are. It applies the same rule to every address.
Why Transparency Matters More Than Efficiency
The standard critique of DeFi is that it is inefficient. It requires overcollateralization. It cannot serve borrowers without crypto assets. It has high gas fees and volatile collateral. These critiques are valid. DeFi is not a replacement for the entire credit system, and it will not be for the foreseeable future.
But the critique misses the point. The value of DeFi is not that it lends more efficiently than a bank. The value is that it reveals what the bank is doing. When a bank denies your mortgage, you get a letter with a vague reason and a phone number. When a DeFi protocol liquidates your position, you can trace the exact price oracle feed, the exact collateral ratio, and the exact block at which the threshold was crossed. The mechanism is not hidden. It is on-chain.
This transparency has a second-order effect that is rarely discussed. It makes the traditional system’s opacity more visible by contrast. When you see a lending protocol where every parameter is public, you begin to ask why your mortgage rate is not. When you see a liquidation that triggers at a mathematically defined threshold, you begin to ask why your credit score denial has no defined threshold at all. The existence of an auditable alternative exposes the arbitrariness of the incumbent. The bank’s black box is no longer invisible. It is just unexplained.
The Cardozo Law Review has documented how DeFi loans were “heavily promoted to the same disadvantaged populations that have experienced exclusion and bias in the traditional system” (Cardozo Law Review, n.d.). This is not a reason to dismiss DeFi. It is a reason to examine it closely. The “black box 3.0” problem—the use of oracles and external data feeds that can introduce new biases—is real (Wiley, 2024). But there is a crucial difference. A bias in a DeFi protocol is a bug in open-source code. It can be identified, reported, and patched. A bias in a bank’s underwriting model is a feature of a proprietary system. It can be denied, litigated, and buried.
The Limits of the Comparison
Intellectual honesty requires acknowledging what DeFi does not solve. Overcollateralization excludes anyone without existing capital. Volatile collateral creates liquidation cascades that can wipe out borrowers during market stress. The governance tokens that control protocol parameters are concentrated in the hands of a few large holders. The oracle problem—the reliance on external data feeds that can be manipulated—is unresolved.
And DeFi is not immune to discrimination in its broader ecosystem. The pseudonymity of blockchain addresses does not prevent exclusionary practices in the fiat on-ramps and off-ramps that connect crypto to the traditional banking system. If you cannot get a bank account, you cannot easily buy the collateral you need to borrow on Aave. The discrimination at the boundary between TradFi and DeFi is as real as the discrimination inside TradFi.
But the comparison is not between a perfect system and an imperfect one. It is between a system that hides its mechanism and a system that exposes it. The traditional lender’s model is proprietary. The DeFi protocol’s model is public. The traditional lender’s denial is opaque. The DeFi liquidation is traceable. The traditional lender’s biases are deniable. The DeFi protocol’s bugs are auditable.
Transparency does not guarantee fairness. But opacity guarantees that fairness cannot be verified. The first step toward accountability is visibility. DeFi provides that visibility. The traditional system does not. And that is why, for all its flaws, the open-source algorithm is superior to the black box—not because it is perfect, but because it can be read.
References
Cardozo Law Review. (n.d.). Crypto-native credit score: Between financial inclusion and predatory lending.
Elzayn, H., Freyaldenhoven, S., Kobler, R., & Shin, M. (2025). Measuring fairness in the U.S. mortgage market. Federal Reserve Bank of Philadelphia Working Paper, WP 25-04.
Federal Reserve. (2022). Comment letter on Community Reinvestment Act regulations. Board of Governors of the Federal Reserve System.
Kindred Futures. (2025). Major drivers of mortgage denial disparities.
Management System Engineering. (2025). Fairness in focus: Quantitative insights into bias within machine learning risk evaluations and established credit models. Management System Engineering, 4(8).
National Fair Housing Alliance. (2025). The state of equitable homeownership 2025 report.
Operation HOPE. (2025, September 8). Closing the gap in Black homeownership.
Proof of Talk. (2026). DeFi lending and vaults: Where the yield comes from.
Wiley. (2024). Decentralized credit scoring: Black box 3.0. ORIGINAL ARTICLE.
Court of Justice of the European Union. (2025). Judgment in Case C-203/22, Dun & Bradstreet Austria. Press Release No 22/25.
European Parliament. (2024). Communication to Members: Petition No 0794/2023 on the transparency of the German credit rating agency Schufa. Committee on Petitions.
Kreditai. (2026). Credit scores don’t cross borders, and that’s costing European borrowers thousands.
Zenodo. (n.d.). The score punished me: Reflexive opacity arms race in algorithmic credit assessment.