Model Risk Management for Large Language Models in Financial Services

This book provides a practitioner-focused framework for managing model risk in large language model applications across regulated financial services. It shows how classical model risk management must be extended for systems that are non-deterministic, composition-based, vendor-dependent and often embedded in consequential customer- or market-facing processes.

The book develops an operational discipline for validating, inventorying, monitoring and governing LLM applications in banks and other regulated firms. It introduces an artefact taxonomy for decomposing LLM systems into governable components, an inventory schema covering prompt templates, retrieval indexes, foundation models and agent tooling, and a fingerprint-bound validation approach for systems whose outputs cannot be reproduced in the classical sense. Chapters cover prompt engineering, third-party model governance, design-time and independent validation, metric portfolios, fairness and bias, Consumer Duty, agent governance, red-teaming, monitoring and incident response.

Sector chapters apply the framework to retail and commercial banking, insurance and capital markets, with use cases including customer correspondence, credit memo drafting, claims classification, research summarisation and regulatory reporting. Accompanied by runnable Python notebooks, it is written for model risk managers, validators, CROs, Chief AI and Data Officers, regulators, data scientists and technology leaders seeking practical governance tools for generative AI.