Ongoing Performance Monitoring for LLM and Agentic AI in Banking: A Validation and Model Risk Handbook: Designing, Validating, and Supervising LLM and ... AI Systems Across the Three Lines of Defense Paperback – March 5, 2026

Your model passed validation. Now what?Validation is a point-in-time opinion. It confirms that a system was sound on the day it was reviewed. It does not tell you what the system is doing today, after silent prompt changes, vendor-side model updates, retrieval misconfigurations, or knowledge bases that have drifted out of regulatory currency.Large language models and agentic AI systems do not fail loudly. They degrade silently. They hallucinate fluently. They regress across updates no one formally approved. Because they are probabilistic, non-deterministic, and often vendor-controlled, traditional model monitoring frameworks cannot reliably detect these failures before they create regulatory exposure.This book closes that gap.Ongoing Performance Monitoring of LLM and Agentic AI Systems presents the first practitioner-level supervision architecture for deployed generative AI in regulated financial institutions.Written for model risk validators, compliance officers, internal auditors, technology owners, and risk leaders operating under SR 11-7, BSA, ECOA, and regulatory reporting requirements, the book provides:A reinterpreted Three Lines of Defense framework for LLM-specific governanceProcess-level risk analysis across customer chatbots, credit underwriting, AML screening, and regulatory reportingA four-dimension Ongoing Performance Monitoring architecture covering output quality, operational performance, governance integrity, and escalationExamination-ready evidence standards aligned with supervisory expectationsEnterprise governance controls for prompt management, observability design, model inventory, vendor oversight, and regression detectionThis is not a book about AI theory. It is a governance operating manual for systems already in production.No technical background is required. Accountability is.Fourteen chapters. Three sections. One framework practitioners can implement, defend, and hand to an examiner. Read more