Background and objective: The overall performance of automated insulin delivery (AID) systems could be affected by various faults, including glucose sensors', insulin pumps', or patient-induced faults. The main objective of this review is to determine the faults that crop up most commonly in clinical trials and to discuss the methods for detecting and mitigating AID faults.Method: PubMed and Google Scholar databases were used to collect the literature. Initially, 206 English manuscripts published in the last six years were selected, of which 38 manuscripts (including 9 clinical trials) were shortlisted for this review.Results: We propose a taxonomy of the AID system focusing on the possible faults and later a detailed analysis of each category (CGM sensor faults, insulin pump faults, and patient-induced faults) was performed.Conclusions: Investigation reveals that connectivity problems, infusion site failures, and infusion set faults are more prevalent in clinical trials. Several model-based and data-driven approaches used in the shortlisted papers for the detection and mitigation of these faults are reviewed to highlight gaps in the study. The review's concluding remarks state that it is essential to emphasize the requirement for safety in future AID systems and that the patient must be regarded as an intrinsic element of the system to develop a fully automated AID system.