Ethical impact assessment should begin early and remain iterative throughout the AI lifecycle, making D correct. Waiting until production, a user threshold, or an actual ethical incident would make the assessment reactive rather than preventive. UNESCO describes ethical impact assessment as an " ethics by design " mechanism covering research, design, development, deployment, use, maintenance, monitoring, and eventual termination, while emphasizing its particular value during early research and design. OECD-listed assessment guidance similarly states that the first assessment should occur during design, with reassessments at development, deployment, and other significant lifecycle stages. NIST ' s AI RMF Playbook also recommends impact assessment at key stages because risks, operating conditions, stakeholders, data, and system capabilities can change over time. Continuous reassessment therefore supports proactive identification and mitigation of emerging harms.