This paper presents a novel approach to real-time firearm detection using advanced YOLO models and a custom dataset derived from the Grand Theft Auto V (GTA V) video game. Our work encompasses three primary contributions: the creation of a unique dataset., the development of an effective detection model, and the implementation of a desktop application for real-time alerting. Firstly, we constructed a comprehensive dataset from GTA V to address the limitations of existing datasets, which often lacked modern scenarios and variety in firearm presentation. This dataset includes 2,300 images categorized by distance, firearm type, lighting conditions, and simulated security camera effects. The images underwent augmentation to reduce model overfitting and improve diversity. Our detection model leverages the YOLO architecture, with extensive experiments comparing YOLOv7, YOLOv8, and YOLOv10. YOLOv8 achieved the highest mean Average Precision (mAP50-95) of 0.70485, significantly outperforming YOLOv7 and YOLOv10. YOLOv7 and YOLOv8 were fine-tuned using weights from pre-trained models and adjusted hyperparameters to optimize performance. Additionally, we developed a desktop application to interface with security camera feeds. The application processes images to detect firearms and notifies operators with both auditory and visual alerts. It records incidents and provides tools for real-time crime detection, enhancing security measures. Our results in real-world simulated situations demonstrate the effectiveness of using a custom video game dataset and state-of-the-art YOLO models for accurate firearm detection in real-time applications. Future work will explore further refinements in detection accuracy and application robustness in diverse real-world environments.