Regulatory Science for AI-Discovered Phytopharmaceuticals: Evidence Standards, Model Documentation, Safety Evaluation, and Approval Pathways

Download PDF

Artificial intelligence is increasingly used to identify, rank, and characterize natural-product candidates, yet AI-originated predictions do not resolve the evidentiary challenges associated with complex phytopharmaceutical materials. This article proposes a regulatory science framework for AI-discovered phytopharmaceuticals that integrates two complementary accountability tracks: phytopharmaceutical evidence for identity, quality, composition, consistency, safety, efficacy, and clinical relevance, and AI evidence for data provenance, documentation, transparency, validation, reproducibility, bias control, and lifecycle governance. The framework approach organizes candidate identity and classification, botanical source authentication, phytochemical profiling, batch consistency, quality control, model records, evidence mapping, mechanistic assessment, safety planning, clinical evidence planning where relevant, regulatory translation, risk management, pharmacovigilance readiness, and model-update oversight. Particular emphasis is placed on distinguishing computational prioritization from experimentally supported mechanism, predictive safety from empirical safety evaluation, regulatory alignment from regulatory approval, and documentation readiness from submission readiness. Approval-pathway considerations are therefore treated as structured questions for product-specific scientific assessment rather than as legal recommendations or predetermined routes. The framework also identifies policy priorities for transparent reporting, interdisciplinary review, evidence traceability, terminology harmonization, and continuing monitoring. Its main contribution is a unified regulatory science architecture that connects AI model accountability with the established quality and safety demands of phytopharmaceutical development while preserving clear boundaries against approval claims, reduced-evidence assumptions, autonomous decision-making, and unsupported clinical or regulatory conclusions.