Quantitative structure–activity relationship (QSAR) models for chemicals have been applied in environmental protection by agencies such as ECHA (European Chemicals Agency) and the US EPA (United States Environmental Protection Agency). QSAR models use variables (descriptors) calculated from molecular structures to predict the activity of chemicals with similar structures. These models provide information on chemical toxicity while reducing time and costs and avoiding animal testing. Advances in computer science and the activities of the OECD, ECHA, and the US EPA have accelerated the development and improvement of QSAR methodology. The OECD has established guidelines for validating QSAR models to regulate their use in health and environmental protection. Early QSAR models were linear and multivariate. Today, most leading models are based on machine learning (ML) and artificial intelligence (AI) algorithms, which we plan to use to improve models for the toxicity, mutagenicity, and degradability of chemicals and pollutants in the environment. Our interdisciplinary team aims to address the shortcomings of the best-known models and provide recommendations to improve the OECD guidelines for the validation of regulatory QSARs. We plan to significantly improve (1) model quality assessment, (2) the application of AI algorithms and consensus modelling, (3) validation procedures in QSAR modelling, and (4) toxicity models. The reliability of the QSAR mutagenicity model will be experimentally verified on a set of chemicals that are constantly present in the environment. The improved QSAR methodology will be used to develop more advanced QSAR models, whose use in environmental and health protection within the EU is regulated and mandatory in cases specified under the REACH Regulation 1907/2006. A final report will be sent to the OECD with a proposal to update the guidelines for the validation of regulatory models for toxicity, mutagenicity, and biodegradability of chemicals.