Sound Practices for Responsible Adoption of Artificial Intelligence (AI): Final Report (Financial Stability Board)

Summary

• The Financial Stability Board (FSB) identified 12 sound practices to help all financial institutions realise the benefits of AI while managing its risks responsibly.p.5
• In a case study, a regional bank's machine learning models proactively identified over 95% of non-performing SME loans at least three months before repayment issues occurred in 2022.p.11
• A medium-sized regional commercial bank's GenAI-enabled AI Approver Assistant achieved over 80% accuracy in financial data extraction and reduced credit proposal processing time from more than a day to 15 minutes.p.12
• An agentic AI fraud detection system helped a large bank reduce fraud losses by over 20% in the first half of the 2026 financial year compared to the same period in 2025.p.14
• An AI platform implemented by a global systemically important bank reduced corporate relationship managers' client engagement preparation time by approximately 50%, while enabling around three times more client dialogue.p.15
• A large insurer eliminated approximately 400,000 manual processing instances in 2025 and leveraged AI code generation tools to write six million lines of code with an 80% developer adoption rate.p.16

Key insights

1. The 12 sound practices are split into four practices focusing on organisation-wide AI governance and eight practices covering the full AI lifecycle from inception to retirement.p.5
2. The board and senior management play a pivotal role in setting the strategic direction, establishing boundaries, and providing oversight so that AI adoption aligns with business strategy and risk appetite.p.20
3. Maintaining accurate documentation and lifecycle records of AI use supports accountability, auditability, verifiability, and effective risk management across the organisation.p.25
4. Financial institutions assess the materiality and risks of AI use cases both at the inception stage and periodically thereafter, adjusting controls dynamically as adoption scales.p.31
5. Agentic AI introduces distinct operational and security risks due to high levels of autonomy, including potential unauthorized or erroneous actions across multi-step execution paths.p.22
6. Effective human oversight requires humans to have real authority, capability, and incentives to intervene, avoiding nominal or automated compliance.p.54