AI Guide For Boards in Singapore

AI GUIDE FOR BOARDS IN SINGAPORE

Section III: Governance and Oversight

Chapter 7: Board-Led AI Governance Directors are not expected to understand the algorithmic mechanics of AI models. ey are, however, expected to understand where AI is deployed across the enterprise, what decisions it inuences, what risks it creates, and who is accountable when outcomes deviate or fail.

governance across the AI lifecycle from ideation to decommissioning.

Structured reporting should support meaningful board challenge and oversight.

As AI increasingly inuences how decisions are made within the organisation, boards have to consider who is accountable in the event of unintended outcomes or errors.

Board-led AI governance must be deliberate and structured. is requires establishing clear lines of accountability, dening strict operating rules, and enforcing disciplined

7.1 AI Governance Framework

A Risk-Tiered Approach to AI Governance can help boards balance innovation with appropriate safeguards. is is a similar approach to managing risks using the enterprise risk management (ERM) framework. e three categories and controls discussed below are meant to be illustrative as dierent organisations and industries might tier the risks associated with these categories dierently. As the scale and scope of the use cases evolve in the organisation, boards should revisit the risk tiering to ensure applicability. Organisations should align their specic risk focus depending on the industry, the lifecycle stage and the purpose. e Illustrative Risk Scenarios serve as a general reference, as board risk committees should develop specics for risks associated with their organisations.

Boards should guide management toward implementing a proportionate and risk- based governance framework for AI that also captures new business value. Not all AI systems carry the same level of risk. Applying a uniform set of governance requirements may slow innovation and reduce organisational agility. Instead, governance controls should be proportionate to the nature and potential impact of each AI system, including: • Sensitivity of the data involved. • Degree of automation. • Level of human oversight. • Potential legal, operational, or reputational consequences of failure or misuse.

57

Made with FlippingBook Digital Proposal Creator