Abstract
Abstract Environmental toxicology must evaluate an expanding range of industrial chemicals, emerging contaminants, transformation products, and complex mixtures. Conventional experimental approaches alone cannot efficiently address the scale and heterogeneity of available chemical, biological, and environmental data. This narrative review examines predictive artificial intelligence applications in pollutant hazard assessment, environmental monitoring, exposure modeling, multi-omics integration, and regulatory decision support. Machine-learning and deep-learning models can integrate chemical descriptors, high-throughput screening data, omics profiles, and environmental sensor measurements to support chemical prioritization and mechanism-informed risk assessment. These approaches may also contribute to New Approach Methodologies and reduce reliance on selected animal-testing procedures. However, their reliability is limited by dataset bias, domain shift, insufficient external validation, restricted transferability across chemical classes and ecosystems, and endpoint-specific regulatory requirements. AI should therefore be considered a complementary component of environmental toxicology rather than a standalone replacement for experimental evidence. Future progress requires standardized and representative datasets, transparent model reporting, independent validation, uncertainty characterization, and clearly defined applicability domains. Graphical abstract