Detecting insurance fraud is a major challenge for insurance companies, as fraud poses a significant threat to their profitability and reputation. Fraudsters adopt sophisticated strategies to fool insurance experts, making fraud detection increasingly complex. Insurers need to be extremely precise when assessing the risks associated with insurance cases. Artificial intelligence (AI) is a promising technology that can help insurers spot fraud in real time, preventing or intercepting suspicious false insurance claims before they are approved. o In this paper, we propose an architectural approach of insurance fraud detection system using deep learning algorithms. This approach uses images of damaged vehicles and accident reports to build a virtual assistant AI for detecting insurance fraud in cars. The virtual assistant is a guide for insurance experts to identify insurance applications with characteristics similar to those of previous fraud cases, highlighting inconsistencies or anomalies, and generating alerts to draw experts’ attention to potentially fraudulent cases.