Title: Decoding AI and Machine Learning in Banking Author: Britta Bohlinger, CFE Published: Fraud Magazine (ACFE), Online Exclusive, 1 July 2024 Format: Featured article in ACFE Fraud Magazine, published article ACFE Tags: Financial Transactions and Fraud Schemes, Banking and Financial Services Summary How are artificial intelligence (AI) and machine learning (ML) used in banking operations, specifically in credit risk modeling, fraud detection, loan decision-making? What data governance challenges entail these technologies?Bohlinger, with a background in investment banking risk management and compliance auditing for a government authority, discusses both the benefits and the risks of AI/ML-driven decision-making in regulated financial services. The article reviews the changes in credit risk modeling due to AI. It addresses issues like scorecard opacity, algorithmic bias (including documented cases of discrimination in mortgage approvals based on postal codes), and the limits of machine learning models trained on historical data when faced with new types of fraud. It highlights the "garbage in, garbage out" issue common in machine learning systems and points out the lack of transparency and explainability as main risks for compliance and oversight. The article discusses regulatory and governance frameworks, including the Basel Committee's principles for risk data aggregation, the NIST AI Risk Management Framework, the EU AI Act (finalised April 2024), the Bank of England's guidance on model risk management, EBA ICT risk guidelines, the SM&CR, and the Monetary Authority of Singapore's Veritas FEAT principles. It also covers regulatory sandboxes in the EU, US, and UK as ways to test innovations in a controlled setting. The article ends with suggestions for Certified Fraud Examiners about ongoing model oversight, ethical deployment standards, and the importance of anticipating changes in AI-enabled fraud rather than just reacting. The author is a CFE, Agile-certified professional, and has been an ACFE member since 2014, with contributions to the NIST AI RMF public working group mentioned in the article.
Britta Bohlinger (Mon,) studied this question.