Bias-Aware and Hallucination-Resistant Financial AI for Credit Scoring

Financial institutions are increasingly exploring Large Language Models (LLMs) for financial document analysis and credit assessment. However, hallucinated financial information and demographic bias can undermine the reliability and fairness of AI-assisted lending decisions. This study proposes an integrated framework for trustworthy financial AI that combines hallucination detection, bias mitigation, and fair credit-risk assessment. The framework uses an LLM to extract structured financial information from source documents, followed by an evidence-verification module that identifies unsupported claims through source matching, semantic verification, numerical consistency checks, and confidence assessment. Verified features are then processed by a separate credit-risk model, while a fairness module evaluates demographic and intersectional disparities in extracted information and credit predictions. The framework will be evaluated using public financial question-answering and credit-risk datasets, supplemented by controlled synthetic financial documents containing known factual and numerical errors. Experiments will examine hallucination detection performance, counterfactual demographic bias, credit-risk prediction quality, and the propagation of incorrect financial information into downstream credit decisions. Evaluation will use claim-level precision and recall, false acceptance rate, AUROC, Brier score, calibration, demographic parity, equal opportunity, equalized odds, and credit-decision flip rate. The study aims to determine whether integrating evidence-based verification and fairness controls can reduce the influence of unsupported financial information while maintaining predictive performance and improving the consistency of AI-assisted credit assessment. Keywords - Trustworthy Financial AI; Large Language Models; Hallucination Detection; Bias Mitigation; Fair Credit Scoring; Credit Risk Assessment; Evidence Verification; Algorithmic Fairness.

Authors

Publication Details

Journal
International Journal of Creative and Open Research in Engineering and Management
Published
2026-11-14
DOI
https://doi.org/10.55041/ijcope.v2i11.056
Primary Topic
Ethics and Social Impacts of AI
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

Bias-Aware and Hallucination-Resistant Financial AI for Credit Scoring

Neelotpal Dey, Ms. Apoorva Singh, Ms. Akarshika Pandey, Mr. Shani Kumar Jaiswal
International Journal of Creative and Open Research in Engineering and Management
Ethics and Social Impacts of AI
article

Bias-Aware and Hallucination-Resistant Financial AI for Credit Scoring

Neelotpal Dey, Ms. Apoorva Singh, Ms. Akarshika Pandey, Mr. Shani Kumar Jaiswal
article en

Abstract

Financial institutions are increasingly exploring Large Language Models (LLMs) for financial document analysis and credit assessment. However, hallucinated financial information and demographic bias can undermine the reliability and fairness of AI-assisted lending decisions. This study proposes an integrated framework for trustworthy financial AI that combines hallucination detection, bias mitigation, and fair credit-risk assessment. The framework uses an LLM to extract structured financial information from source documents, followed by an evidence-verification module that identifies unsupported claims through source matching, semantic verification, numerical consistency checks, and confidence assessment. Verified features are then processed by a separate credit-risk model, while a fairness module evaluates demographic and intersectional disparities in extracted information and credit predictions. The framework will be evaluated using public financial question-answering and credit-risk datasets, supplemented by controlled synthetic financial documents containing known factual and numerical errors. Experiments will examine hallucination detection performance, counterfactual demographic bias, credit-risk prediction quality, and the propagation of incorrect financial information into downstream credit decisions. Evaluation will use claim-level precision and recall, false acceptance rate, AUROC, Brier score, calibration, demographic parity, equal opportunity, equalized odds, and credit-decision flip rate. The study aims to determine whether integrating evidence-based verification and fairness controls can reduce the influence of unsupported financial information while maintaining predictive performance and improving the consistency of AI-assisted credit assessment. Keywords - Trustworthy Financial AI; Large Language Models; Hallucination Detection; Bias Mitigation; Fair Credit Scoring; Credit Risk Assessment; Evidence Verification; Algorithmic Fairness.

International Journal of Creative and Open Research in Engineering and ManagementVol. 02(11)
Ethics and Social Impacts of AI
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon