A systematic review was conducted to evaluate the use of artificial intelligence (AI) in antimicrobial stewardship programs (ASPs) and to summarize the predictive performance of machine learning (ML) algorithms compared to clinical decisions in inpatients and outpatients requiring antimicrobial prescriptions. Eighteen observational studies were included from PubMed, Scopus, and Web of Science, focusing on AI applications in ASPs. The studies were excluded if they were in vitro, did not address infectious diseases, or did not reference AI models as predictors. The most commonly used ML algorithms were logistic regression, random forest, support vector machine, and k-nearest neighbors. The most frequently used performance metrics were AUC, sensitivity, specificity, and precision. ML algorithms were found to be useful in identifying inappropriate prescribing practices, selecting appropriate antibiotic therapy, and predicting antimicrobial resistance. The highest AUC value was achieved by the multilayer perceptron, while the gradient boosted tree showed the highest precision in antibiotic selection. Despite the potential benefits of AI in ASPs, there are risks and ethical concerns, including the possibility of biased results due to uncontrolled data inputs and the "black box" nature of AI algorithms. The review highlights the need for further research and the development of standardized tools for quality assessment of ML models. AI can play a positive role in ASPs by assisting in decision-making, but its implementation requires careful consideration of ethical and practical challenges. The findings suggest that AI can be a valuable tool in antimicrobial stewardship, but more research is needed to ensure its effective and ethical application.A systematic review was conducted to evaluate the use of artificial intelligence (AI) in antimicrobial stewardship programs (ASPs) and to summarize the predictive performance of machine learning (ML) algorithms compared to clinical decisions in inpatients and outpatients requiring antimicrobial prescriptions. Eighteen observational studies were included from PubMed, Scopus, and Web of Science, focusing on AI applications in ASPs. The studies were excluded if they were in vitro, did not address infectious diseases, or did not reference AI models as predictors. The most commonly used ML algorithms were logistic regression, random forest, support vector machine, and k-nearest neighbors. The most frequently used performance metrics were AUC, sensitivity, specificity, and precision. ML algorithms were found to be useful in identifying inappropriate prescribing practices, selecting appropriate antibiotic therapy, and predicting antimicrobial resistance. The highest AUC value was achieved by the multilayer perceptron, while the gradient boosted tree showed the highest precision in antibiotic selection. Despite the potential benefits of AI in ASPs, there are risks and ethical concerns, including the possibility of biased results due to uncontrolled data inputs and the "black box" nature of AI algorithms. The review highlights the need for further research and the development of standardized tools for quality assessment of ML models. AI can play a positive role in ASPs by assisting in decision-making, but its implementation requires careful consideration of ethical and practical challenges. The findings suggest that AI can be a valuable tool in antimicrobial stewardship, but more research is needed to ensure its effective and ethical application.