FSI Occasional Papers No 24
The growing use of AI in finance is reshaping operations, risk management, and customer services, but limited explainability of complex models (like deep learning and LLMs) poses serious challenges for institutions and regulators. Explainability is vital for transparency, accountability, compliance, and trust, yet current techniques are often unreliable.
Most existing model risk management (MRM) guidelines are high-level, not tailored for advanced AI, and rarely mention explainability explicitly, making compliance difficult—especially when third-party models are used.
As AI adoption expands in critical business areas, regulators must adapt MRM practices, balancing explainability vs. performance trade-offs. Complex AI could still be used if risks are properly managed, safeguards applied, and restrictions imposed for sensitive use cases (e.g., regulatory capital). Regulators will also need to upskill staff to assess AI effectively.
