AI Policy for Clinical Trial Evidence: Governing Protocol Ambiguity in Simulation

AI policy for clinical trials should help reviewers identify consequential assumptions, assign authority for unresolved choices and determine when evidence needs reassessment. This paper proposes a report, clarify or defer framework that separates interpretation coverage, decision stability and predictive validity. Finite decision identification makes clarification needs checkable: in a constructed longitudinal study, 142 of 144 held-out fixed-threshold cases are already stable. A conjunction counterexample shows how irrelevant questions mislead a query heuristic; a restriction argument proves that deleting readings can lower optimal query cost without establishing completeness. An exploratory public MRI study of 384 patients illustrates the separate model-validity problem. After a retransformation correction chosen following initial test inspection, all 16 cohort forecasts exceed a fixed 0.90 response threshold while two realized summaries do not. This near-threshold example has no untouched confirmatory cohort. Nine located cases across four full protocols distinguish missing context from delegated authority. The policy proposal builds on existing reporting instruments; an AI Governance Command Center binds claims to versioned evidence, unresolved choices and responsible review. Applications concern protocol clarification, amendment review and evidence handoffs, with patient representation and oversight explicit. Faster review, clinical benefit and regulatory acceptance remain untested. Agreement and clarification efficiency are insufficient evidence for interpretation coverage or observed predictive validity.