After starting their AI journeys deploying a handful of models, many banks now have hundreds—sometimes thousands. Each business unit—credit scores, fraud detection, customer churn, pricing, compliance, etc. etc.— builds its own models. And each vendor brings embedded AI. Effective portfolio management means knowing what’s running, where and on what data so you don’t drown in operational overhead or accumulate catastrophic risk. Panelists discuss building centralized model registries that actually get used, implementing continuous monitoring to catch drift before it causes failures, the risk of vendors embedding AI you can’t audit, and why "shadow AI"—models built outside official channels—represents both your biggest risk and best source of innovation, and how to govern it without killing it.