AIGF-1.1 | Risk Management and Internal Controls | 89 |
AIGF-1.2 | AI Ethics Governance Body | 15 |
AIGF-1.3 | Data Management | 61 |
AIGF-1.4 | Algorithm Design and Training | 0 |
AIGF-2.1 | Human-in-the-Loop | 1 |
AIGF-2.2 | Human-on-the-Loop | 1 |
AIGF-2.3 | Human-out-of-the-Loop | 1 |
AIGF-2.4 | Risk-Severity Assessment | 1 |
AIGF-3.1 | Minimising Bias in Data | 1 |
AIGF-3.2 | Explainability | 15 |
AIGF-3.3 | Repeatability and Traceability | 15 |
AIGF-3.4 | Regular Tuning and Monitoring | 1 |
AIGF-4.1 | General Transparency | 1 |
AIGF-4.2 | Accessible Communication | 0 |
AIGF-4.3 | Feedback Mechanisms | 1 |
AIGF-4.4 | Disclosure of AI Use | 1 |
SGAIGOV-1 | Internal Governance Structures and AI Ethics | 0 |
SGAIGOV-2 | Risk Assessment and Determining Human Involvement | 0 |
SGAIGOV-3 | Operations Management - Data, Algorithms, Robustness | 0 |
SGAIGOV-4 | Stakeholder Interaction, Communication, Transparency | 0 |
SGAIGOV-5 | GenAI Safe Use, Provenance, Watermarking | 0 |
AGT-2.1.1 | Select agent use cases by assessing impact and likelihood factors and threat modelling | 0 |
AGT-2.1.2 | Bound agents by design: least-privilege tools, SOP-constrained autonomy and contained impact | 0 |
AGT-2.1.2-ID | Give every agent a unique, accountable, catalogued identity and scoped, bounded authority | 0 |
AGT-2.2.1 | Allocate accountability across the agent value chain, internally and in contracts | 0 |
AGT-2.2.2 | Require human approval at significant checkpoints and audit that oversight stays effective | 0 |
AGT-2.3.1 | Prefer deterministic system-level controls over prompt guardrails, with runtime controls | 0 |
AGT-2.3.2 | Test agents end to end before deployment, individually and as a system, repeatedly and realistically | 0 |
AGT-2.3.3 | Roll out gradually, monitor and log in layers with defined interventions, and manage change | 0 |
AGT-2.4 | Tell users they face an agent and what it can do, and train integrating users to oversee it and keep their skills | 0 |
FAMILY | Singapore's model AI governance frameworks: what they are | 0 |
GEN-1 | Allocate accountability across the generative AI chain upfront and provide safety nets | 0 |
GEN-2 | Use trusted, quality training data and handle personal data and copyright pragmatically | 0 |
GEN-3 | Apply baseline safety practices, evaluate models and publish food-label disclosures | 0 |
GEN-4 | Run vulnerability reporting and AI incident reporting with a materiality threshold | 0 |
GEN-5 | Use independent, standardised third-party testing and assurance | 0 |
GEN-6 | Adapt security-by-design to generative AI threats with input filters and forensics | 0 |
GEN-7 | Label AI-generated content with watermarking or cryptographic provenance where practical | 0 |
GEN-8 | Support safety and alignment research and AI for public good | 0 |
MAF-3.14 | Choose human-in, over- or out-of-the-loop oversight from the probability and severity of harm | 0 |
MAF-3.23 | Practise data accountability: lineage, quality, bias, separate datasets and periodic review | 0 |
MAF-3.27 | Make models explainable, or document repeatability where explanation is not feasible | 0 |
MAF-3.32 | Test robustness and tune models regularly in conditions like production | 0 |
MAF-3.37 | Build traceability, reproducibility and auditability in proportion to risk | 0 |
MAF-3.46 | Disclose AI use and set a policy for which explanations are given and when | 0 |
MAF-3.52 | Offer opt-out or recourse, feedback and human review of material AI decisions | 0 |
MAF-3.55 | Test interfaces, communicate plainly, set acceptable use policies and get information from AI vendors | 0 |
MAF-3.7(1) | Allocate clear, trained and resourced roles for ethical AI deployment | 0 |
MAF-3.7(2) | Run risk management and internal controls specific to the deployed model | 0 |
MAF-3.9 | Weigh commercial objectives against AI risks, guided by corporate values, in a reviewed risk impact assessment | 0 |
STATUS | Status: voluntary guidance, current texts MAF 2020, GenAI 2024 and Agentic v1.5 (2026) | 0 |