Iran’s National Artificial Intelligence Development Act introduces a six-month cycle for evaluating domestic and foreign AI systems by risk level, yet it does not determine the legal effects of a later reclassification. This article examines the consequences of an increase or decrease in risk for future deployment, ongoing use, and administrative decisions already made with the system’s support. It combines doctrinal analysis of Iranian law on temporal effects and revision of administrative decisions with a functional comparison of the EU AI Act and lifecycle-oriented risk-management frameworks. The article argues that reclassification has no single legal character: it may disclose a risk that already existed, result from a later change in the model, data or context of use, or follow from a change in the regulatory standard itself. That distinction matters for temporal effect. A new classification should govern future requirements and trigger a reasoned, proportionate reassessment of ongoing deployment. For prior decisions, the new assessment should operate first as evidence rather than as an automatic ground of invalidity. Reopening is justified only where the risk existed at the relevant time, concerned the function actually used, and was capable of materially affecting the procedure, right, or outcome; the corrective measure must then be proportionate. The proposed four-stage test—temporal relevance, functional relevance, material effect, and proportionality of corrective action—seeks to preserve legal certainty while enabling meaningful review where later evidence reveals a genuine defect in earlier administrative decision-making.