Artificial Intelligence for Systemic-Risk Monitoring in the U.S. Financial System: A Narrative Review of Methods, Data Sources, and Supervisory Applications

Systemic-risk monitoring in the United States was reorganized after the 2007-09 financial crisis, and artificial intelligence (AI) is now becoming part of that supervisory landscape.This narrative review examines how AI and machine learning bear on U.S. systemic-risk monitoring.Sources spanned peer-reviewed methodological and empirical studies, primary U.S. regulatory and supervisory documents, and policy reports.International studies served as comparators where U.S.specific evidence was thin.This narrative review addresses three questions: which methods and data sources support systemic-risk monitoring, under what conditions machine learning improves early-warning prediction relative to conventional statistical models, and what limits the supervisory use of AI-based monitoring.Evidence assembled here describes a developed post-crisis architecture built around financial-stability oversight, market-based systemic-risk measures, stress-testing tools, supervisory data collections, and emerging AI governance frameworks.Evidence for machine learning-based early warning is promising but uneven.Reported performance varies with jurisdiction, crisis definition, sample period, data source, and evaluation design, and many favorable findings come from cross-country or non-U.S.settings.Data infrastructure is the strongest recurring limitation.Coverage, entity identification, linkage across institutions and markets, interoperability, and reporting frequency shape what both conventional and AI-enabled monitoring systems can observe.This review contributes a U.S.-focused synthesis connecting systemic-risk measurement, supervisory data infrastructure, and AI governance, and it concludes that reliable AI-assisted monitoring depends not only on model development but also on more standardized, linkable, and timely supervisory data.