Artificial intelligence revolution in toxicology: Clinical precision, global equity, and the 2030 roadmap

Artificial intelligence (AI) is transforming toxicology by enabling faster, more accurate, and more equitable approaches to diagnosis, treatment, research, and education. This review synthesizes recent advances in AI applications across clinical toxicology, predictive modeling, pharmacovigilance, and training, highlighting innovations such as deep learning for diagnostic stratification, reinforcement learning for personalized antidote dosing, and generative AI for virtual patient simulations. While these technologies demonstrate substantial promise, their clinical and regulatory adoption remains constrained by algorithmic bias, limited model interpretability, validation challenges specific to feature importance accuracy, and persistent global digital inequities. Critical limitations include the distinction between target prediction accuracy and feature importance accuracy in supervised models, where high predictive performance does not guarantee mechanistic reliability. Moreover, the potential for AI-generated misinformation and the need for continuous human oversight in clinical contexts warrant careful consideration. To guide responsible integration, the "ToxAI Pact" is proposed, a 2030 roadmap emphasizing harmonized validation standards, robust feature importance validation protocols, watermarking of generative outputs, and infrastructure investment for low-resource settings. By embedding fairness, explainability, and robust governance, AI can evolve from experimental tools into foundational infrastructure for safer, more inclusive toxicology worldwide.