Singapore AI Governance Framework MAF-3.14: Choose human-in, over- or out-of-the-loop oversight from the probability and severity of harm

What does MAF-3.14 require?

MAF paragraphs 3.14 to 3.18: classify oversight as human-in-the-loop (the human decides; AI only recommends and must give enough information), human-over-the-loop (the human monitors and can take control or adjust parameters) or human-out-of-the-loop (no human override); choose using a probability-severity of harm matrix defined for the context, also weighing whether harm is physical or intangible, whether it can be undone, and whether human involvement is operationally feasible; for safety-critical systems let a person assume control or safely shut down the system; review the choice regularly as norms evolve.