Six Core Principles
Anchored in MAS MindForge and mapped across 14 regulatory frameworks — a unified taxonomy for AI risk governance in Singapore financial services.
Clear Human Accountability and Oversight
Defined ownership, roles, and escalation paths for AI systems must exist at every level — from board to operational team. Governance documents institutionalise accountability. Humans remain ultimately responsible for AI outcomes; oversight structures are designed to make that responsibility meaningful, not nominal.
Proportionate, Risk-Based AI Governance
Governance depth and controls are calibrated to the materiality and inherent risk of each AI use case. Enterprise risk appetite, KRIs, third-party AI risk management, and use-case- level risk assessment prevent both under-governance and box-ticking compliance. Proportionality is a first-class principle, not an afterthought.
Responsible Use Across the AI Lifecycle
Controls apply at every stage from design through decommissioning — not just at deployment. Includes use-case context alignment, third-party onboarding, pre-deployment testing, deployment planning, ongoing monitoring, and change management. Independent validation is a critical control embedded at multiple lifecycle stages.
Data, Model, and System Integrity & Soundness
AI systems must be technically trustworthy: training data is ethical, high-quality, and representative; models are validated against bias and fairness standards; guardrails and explainability measures (at the model level) are in place; and infrastructure is fit for purpose. Note on fairness: MindForge operationalises fairness as a data-and-model-integrity obligation (C8 ethical data use, C9 bias controls) rather than as a separate governance pillar. MAS FEAT's fairness requirements are therefore mapped here. This is consistent with treating fairness as a technical property of the model and data pipeline, not a standalone policy category.
Transparency, Traceability, and Auditability
AI decisions must be explainable at the decision level, systems must be inventoried, documentation must be sufficient for independent review, and change logs must enable retrospective audit. Covers both MAS FEAT transparency obligations (decision-level explainability) and EU AI Act Articles 11–13 (technical documentation, record-keeping, and disclosure to deployers).
Organisational Capability and Responsible AI Culture
Governance is only as durable as the people and infrastructure sustaining it. Requires AI literacy, role-specific training, interdisciplinary representation in governance bodies, and infrastructure readiness. Addresses the sustainability of the governance framework over time — preventing governance decay as models proliferate and teams turn over.
How Each Principle Maps Across Frameworks
Click a principle row to highlight its coverage. Hover a cell for alignment strength detail.