The Financial Stability Board (FSB) has published 12 recommended practices to help financial institutions adopt artificial intelligence responsibly, amid growing concerns over the operational, cyber security and financial stability risks associated with the technology.

Published today (15 October), the report, Sound Practices for Responsible Adoption of Artificial Intelligence, examines how financial institutions can harness AI’s benefits while maintaining effective oversight and risk management.

The FSB highlights particular concerns surrounding generative and agentic AI, warning that increasingly autonomous systems could act unpredictably, take unauthorised actions or disrupt connected infrastructure.

One case study involving a large life insurance company found that, during testing, AI agents could be manipulated into attempting to access unauthorised systems, with one attempting to delete data files.

The report also identifies risks associated with AI-enabled cyberattacks, opaque decision-making, third-party technology dependencies and workforce deskilling.

Against this backdrop, the FSB has outlined 12 voluntary practices, divided into four recommendations on organisation-wide governance and eight covering the management of AI throughout its lifecycle.

The 12 Recommended Practices

1. Strategic direction and oversight: Ensure AI adoption aligns with the organisation’s business strategy, risk appetite and objectives, with appropriate oversight from senior leadership.
2. Governance and accountability: Establish clear responsibilities and governance structures for managing AI adoption and its associated risks.
3. AI risk management: Integrate AI-specific risks into existing risk management frameworks, with appropriate documentation, assessments and records of AI use.
4. Organisational adaptability: Continuously review and update governance, capabilities and risk management practices as AI technologies evolve.
5. Risk assessment: Assess the potential risks and significance of AI applications before deployment and throughout their use.
6. Technology selection: Choose AI models and systems based on business requirements, technical suitability and the risks associated with their intended applications.
7. Data governance: Maintain accurate, reliable, complete and secure data for AI training, testing and deployment.
8. Explainability and transparency: Understand how AI systems produce outcomes, favour explainable models where appropriate and introduce additional controls where transparency is limited.
9. Performance management: Regularly test, assess and monitor AI systems to ensure they continue to perform reliably and appropriately.
10. Human oversight: Maintain effective human supervision proportionate to the risks, complexity and autonomy of individual AI applications.
11. Cyber and ICT security: Address AI-specific cyber threats through risk management, testing, information-sharing and appropriate security controls.
12. Third-party risk management: Manage dependencies on external AI providers, considering performance, transparency, data quality, supply chain concentration and business continuity.

The FSB emphasises that the practices are not binding international standards, but a flexible framework that financial institutions can adapt according to their size, complexity and exposure to AI-related risks.

Recommended reading

• New Scottish Centre of Excellence For Digital Trust Launches
• Businesses Given One Month to Meet New ICO Data Protection Complaints Rules
• How Important Are Digital Trust Frameworks to Firms?
• DIGIT Expo 2025 | Is Digital Consumer Trust on The Brink?

More extensive safeguards may be appropriate for larger institutions or those deploying AI in critical functions, while smaller organisations and lower-risk applications may require more proportionate approaches.

The final report follows a public consultation launched in June 2026, with revisions incorporating additional considerations around agentic AI, documentation, transparency and third-party risks.

The FSB said it will continue monitoring developments and update its recommended practices as AI technologies and their applications evolve.