The Financial Stability Board published final guidance urging financial institutions to govern AI organization-wide and manage risks throughout its lifecycle.
The Financial Stability Board (FSB) published its final report on responsible AI adoption in finance on October 15, 2026, concluding a public consultation that began in June. The document sets out 12 “sound practices” for financial institutions and supervisors; it is presented as a practical toolkit, not a new binding rulebook.
The recommendations cover two broad areas. One calls for organization-wide governance, with boards and senior managers overseeing how AI use fits a firm’s business model, strategy and risk appetite. It also emphasizes clear responsibilities, documenting AI use, incorporating AI risks into existing risk-management frameworks and adapting controls as systems change.
The second area follows AI across its lifecycle, from development through deployment. Among the practices are assessing use-case risks, selecting appropriate technologies, managing data, evaluating performance and providing human oversight. The report also addresses explainability, AI-related cyber and information-technology risks, and dependencies on third-party providers. It includes case studies describing how financial institutions have applied practices proportionately.
The guidance matters because financial firms may use AI in operationally important and customer-facing workflows, while relying on external models, data and infrastructure. The FSB says the practices are intended to help firms manage those risks and support financial-system resilience, while encouraging coordination and information-sharing among institutions and authorities across jurisdictions.
The final report reflects feedback on the June consultation draft, and the FSB also published an overview of consultation responses. It encourages boards and senior management to use the practices when developing or refining governance, risk-management and control frameworks. The organization says it will monitor AI developments and update the toolkit if needed; the published recommendations do not themselves establish a uniform legal obligation for every institution.