This paper examines how modern machine-learning techniques are reshaping consumer credit scoring and what legal-regulatory guardrails are required to keep accuracy accountable. It argues that, in tabular lending contexts, ensemble models (e.g., gradient-boosted trees) deliver meaningful lifts over traditional logistics while remaining compatible with faithful, applicant-level explanations (e.g., SHAP). The core risks are opacity, drift, and disparate impacts—problems that cannot be solved by technical fixes alone but must be governed through clear model-risk controls, documented fairness choices, auditable data lineage, and reason-code specificity. Surveying leading regimes, the paper highlights the EU AI Act’s “high-risk” obligations for creditworthiness assessment, U.S. expectations under ECOA/Reg B and SR 11-7/OCC model-risk guidance, MAS FEAT/Veritas operationalization, and BIS/BCBS prudential concerns about correlated model failures. Against this backdrop, Uzbekistan’s existing statutes on personal data, consumer credit, and credit histories provide an enabling legal scaffold. The paper proposes targeted secondary rules—explainability standards tied to actual computations, ex-ante fairness metrics and monitoring, mandatory drift detection with challenger models, and full vendor oversight—to align innovation with due process. Done well, AI scoring can expand inclusion, strengthen dignity through actionable explanations, and enhance system resilience without trading off consumer protection.

Adverse-action reasons; Basel/BCBS prudential risk; Consumer protection; Credit histories; Data governance; Drift monitoring; ECOA/Reg B; EU AI Act (high-risk); Explainability (SHAP); Fairness metrics (equal opportunity); Gradient-boosted trees; Model calibration; Model risk management (SR 11-7/OCC); Monotonic constraints; Personal data law (Uzbekistan); Tabular ML; Transparency & accountability; Vendor model oversight; Veritas/FEAT (MAS); Uzbekistan—consumer credit law.

EU AI Act—creditworthiness as “high-risk” (European Commission/Parliament texts); EBA Big Data & Advanced Analytics; U.S. SR 11-7 (Federal Reserve) and OCC Model Risk Management; CFPB guidance on adverse-action specificity; MAS FEAT principles; BCBS/BIS notes on AI prudential risk; Uzbekistan: Law on Personal Data; Law on Consumer Credit; Law on Exchange of Credit Information (all via Lex.uz). ↩

Systematic reviews of ML in credit risk (e.g., Dastile et al., Applied Soft Computing, 2020); survey papers on ML for default prediction; Rudin, “Stop explaining black box models…” (Nature Machine Intelligence, 2019); SHAP explainability for tabular credit models. ↩

Hardt, Price, & Srebro (2016) on equal opportunity; Kleinberg, Mullainathan, & Raghavan (2016) on inherent trade-offs; additional industry guidance on fairness testing in lending. ↩

EU AI Act (obligations for high-risk systems); EBA Guidelines on Loan Origination & Monitoring; EBA report on Big Data/Advanced Analytics—governance, data quality, validation expectations. ↩

CFPB circulars/press materials clarifying adverse-action specificity for AI-driven denials under ECOA/Reg B; FRB SR 11-7 on model risk; OCC Comptroller’s Handbook sections on validation and independent challenge. ↩

MAS FEAT principles and Veritas initiative (tooling for responsible AI in finance); BIS/BCBS speeches and notes warning about systemic risks from opaque, correlated ML models. ↩

Republic of Uzbekistan: Law “On Personal Data” (ZRU-547); Law “On Consumer Credit” (as amended); Law “On Exchange of Credit Information (Credit Histories)” (ZRU-301)—all published on Lex.uz. ↩

High-impact classification logic inspired by EU AI Act; institution-wide controls per SR 11-7/OCC; CFPB’s expectation of faithful, specific adverse-action reasons; EBA/MAS guidance on documentation and transparency. ↩

Practical MLOps/validation practices aligned with EBA documentation norms and OCC/FRB expectations: model inventories, lineage, challenger frameworks, calibration/fairness dashboards. ↩

CFPB specificity as a guardrail against “template reasons”; SR 11-7 on independent validation; literature on proxy variables and sensitivity testing for fairness in credit. ↩

BCBS prudential framing (AI and systemic resilience); MAS/EBA transparency for consumer trust; CFPB consumer-rights benefits; accountability regime in EU AI Act. ↩

Synthesis of Uzbek legal anchors (Lex.uz) and international supervisory materials; academic sources on explainability and fairness cited above. ↩