FSB finalises sound practices for responsible AI adoption
The Financial Stability Board (FSB) has published the final version of its sound practices for financial institutions adopting artificial intelligence.
The 12 practices cover organisation-wide governance and the management of risk across the AI lifecycle, and the FSB strongly encourages boards and senior management to refer to them when making decisions on strategy, technology and risk.
Compared with the consultation draft, the final report says more about the benefits and risks of AI, including workforce de-skilling. It adds guidance on documentation for auditability and traceability, clarifies the roles of boards and senior management at smaller institutions, broadens its treatment of third-party AI risk and introduces considerations for agentic AI and multi-agent systems.
Data governance remains Sound Practice 7, which asks institutions to keep the data used to train, test and run AI fit for purpose: accurate, complete, consistent, reliable and secure. The report warns that weak data governance and poor-quality data can lead to biased outputs, performance problems and legal breaches, and its documentation guidance covers the data sources used during inference, as well as training data for models built in-house. Data quality also features among the risks to manage in third-party AI dependencies.
Anyone building AI on top of reported data should take the documentation point to heart. Tracing an output back to its inputs is far easier when each figure carries its own context, and tagged regulatory data arrives with its entity, period, unit and a defined concept attached.
The FSB will monitor developments and update the practices as needed.
Read more here.
