What does IMDA Model AI Governance Framework for Generative AI say about AI in banking?

What are the key points of IMDA Model AI Governance Framework for Generative AI?

• Nine dimensions: accountability, data, trusted development and deployment, incident reporting, testing and assurance, security, content provenance, safety and alignment research and development, and AI for public good.
• Accountability: responsibility should be allocated along a layered development chain, drawing parallels with 'today’s cloud and software development stacks' — relevant to banks relying on foundation-model vendors.
• Trusted development and deployment calls for industry best practice in development and evaluation and 'food label'-type transparency and disclosure.
• Incident reporting: 'no software we use today is completely foolproof'; the framework calls for structures to monitor and report AI incidents, noting incident reporting is established practice in critical domains such as finance.
• Testing and assurance: third-party testing plays a complementary role, as it does today 'in many domains, such as finance and healthcare'.
• Security: generative AI 'introduces the potential for new threat vectors against the models themselves', so existing information-security frameworks need adapting.
• It builds on the 2020 Model AI Governance Framework for traditional AI, whose guidance is incorporated under the trusted development and deployment dimension.
• MAS's Guidelines on AI Risk Management also point banks to IMDA's Starter Kit for Testing LLM-Based Applications for Safety and Reliability and its Transparency Guidelines for Generative AI Chatbots.

What did IMDA Model AI Governance Framework for Generative AI change for banks?

The 2020 framework addressed organisations deploying traditional, predictive AI; the 2024 framework widened the lens to foundation models and the ecosystem around them, adding incident reporting, content provenance and third-party testing. Banks are not its primary audience, but MAS's AI Guidelines direct institutions to IMDA's frameworks and testing resources as complements to MAS's own expectations.