Exploring Explainable AI in the Financial Sector: Perspectives of Banks and Supervisory Authorities

Supervisory authorities and banks have conflicting views on how much AI systems need to be explained: The study found a "disparity" between what regulators want and what financial institutions think is necessary regarding AI explainability, suggesting the financial sector lacks alignment on this critical issue.

Different types of explainability are being confused: The researchers argue that the sector conflates "technical AI (model) explainability" with broader "AI system explainability," indicating that clearer definitions and differentiation between these concepts could help resolve current disagreements and improve regulatory clarity.

Explainability matters most in high-stakes financial decisions: The three use cases examined—consumer credit decisions, credit risk assessment, and anti-money laundering detection—were chosen because they directly affect people's access to financial services and regulatory compliance, highlighting where transparency is most critical.

Current regulations may not adequately address AI transparency needs: The study's focus on "applicable laws and regulations" suggests that existing legal frameworks may not provide sufficient guidance on what explainability actually means in practice for AI systems used in finance, leaving both banks and regulators uncertain about compliance requirements.