AI Guide For Boards in Singapore

AI GUIDE FOR BOARDS IN SINGAPORE

Section III: Governance and Oversight

e Board Checklist for AI Governance serves as a practical diagnostic tool for directors to evaluate the maturity of their organisation’s AI governance framework.

Boards should consider if AI is creating value in a form the organisation can sustain, explain, govern and defend.

Board Checklist for AI Governance

Strategy and risk appetite • Has the board formally reviewed and updated the Risk Appetite Statement to include explicit boundary conditions and zero-tolerance thresholds for AI deployment? • Is there a clear, board-approved policy dening the maximum acceptable level of technical and governance debt regarding unexplainable or legacy AI systems? • Does the company's long-term corporate strategy explicitly address potential competitive displacement or disruption by AI-native business models? Roles and accountabilities • Is there a single, named senior management executive (e.g., a Chief AI Ocer) clearly accountable for managing AI risk and unlocking AI value across the entire enterprise? • Are the boundaries of oversight between the Risk, Audit, and Technology committees clearly mapped out to eliminate oversight gaps or duplicate eorts? • Does the internal audit function possess the specialised data science and algorithmic auditing capabilities required to independently validate high-risk models? Frameworks and operational routines • Is there a dynamic, regularly updated enterprise AI inventory that categorises all active and in-development models by their risk tiers? • Does every Tier-3 (High Risk) AI deployment have an explicitly assigned business owner, a documented lineage of training data, and regular validation routines? • Are third-party AI vendors bound by strict contractual clauses regarding model dri notications, data privacy guarantees, and independent verication rights? Operational resilience and recovery • Has management developed and thoroughly tested functional “algorithmic kill- switches” and deterministic fallback modes for all core automated processes? • Is there a dened communication and operational playbook for responding to AI- specic crises, such as public bias accusations, data poisoning, or model hallucinations? Reporting and metrics • Does the board receive a regular, trend-focused risk report featuring quantitative Key Risk Indicators for model stability, dri, and human override rates? • Are management’s nancial claims regarding AI productivity or eciency gains calculated against full cost baselines, including hidden human-in-the-loop and data prep costs? • Is there an established regulatory horizon scanning process that actively updates the board on shiing compliance requirements across our operating markets?

71

Made with FlippingBook Digital Proposal Creator