From Trolley Problems to Contextual Cascades: Rethinking Ethical Context in AI Governance

Artificial intelligence is increasingly embedded in extended institutional processes rather than isolated decisions. Existing work has shown why ethical evaluation must attend to context, sociotechnical abstraction, distributed responsibility, and feedback. This article develops a narrower normative process account of what happens when an intervention at one stage changes the conditions under which subsequent decisions are made. It introduces Contextual Cascades, centred on normative reconstitution: a sociotechnical transformation is normatively reconstitutive when it materially changes a normatively relevant relation—such as standing, evidentiary position, vulnerability or power, duties or entitlements, or feasible and justifiable options—and that altered relation conditions a subsequent decision or response. The framework distinguishes such transitions from mere complexity, data transformation, or causal feedback through three conditions: stage dependence, normative transformation, and propagation. A normative conceptual research design is combined with structured documentary reconstruction of Australia’s Robodebt scheme and the Dutch System Risk Indication (SyRI). The cases show how representations can acquire institutional force through different administrative pathways and how corrective reasoning may propagate or be contained. The article derives three governance requirements: context recognition, context preservation and recoverability, and contextual accountability. It argues that AI governance must assess not only individual outputs, but also how sociotechnical processes construct, transform, and propagate the normative conditions of later decisions. Preprint prepared in connection with Data for Policy 2026. This version has not undergone peer review.