The PFAS Nexus: A Critical Review Linking Toxicity, Remediation, and Emerging Artificial Intelligence Applications

Per- and polyfluoroalkyl substances (PFAS) comprise a structurally heterogeneous class whose reported size varies with the operational definition and inventory used. Their environmental persistence, mobility, and widespread occurrence have created substantial challenges for exposure assessment, toxicological characterization, and remediation. This review critically integrates three connected domains: molecular and systems toxicology, destructive and non-destructive remediation technologies, and emerging artificial intelligence and machine-learning (AI/ML) applications. The toxicology section emphasizes the strength of evidence across metabolic, endocrine, hepatic, immune, reproductive, epigenetic, and carcinogenic outcomes, while avoiding extrapolation from a small number of well-studied legacy compounds to the full PFAS chemical space. Remediation approaches are compared by contaminant range, treatment outcome, evidence for defluorination or mineralization, relative energy demand, residual generation, and major implementation limitations, with expanded coverage of non-thermal plasma, fungal biotransformation, microbial consortia, and hybrid physicochemical-biological treatment. For AI/ML, we distinguish PFAS-specific demonstrations from methods transferred from adjacent fields and from longer-term research concepts that remain unvalidated. We define the PFAS Nexus as a conceptual decision framework that links hazard evidence, treatment selection, data quality, and regulatory needs rather than as a validated predictive model. The synthesis identifies priority needs for transparent datasets, external model validation, mass-balance and fluoride-recovery measurements, matrix-realistic biological studies, and pilot-scale evaluation of integrated treatment trains.