​Search and rescue (SAR) operations are often hindered by limited coverage, slow response times, and operational risks, making rapid and reliable victim detection a critical challenge. To address these limitations, this study presents an AI-driven UAV framework that integrates a simulated multi-UAV routing with a YOLOv8-based human detection model. A region-specific aerial dataset consisting of 2430 images and 2831 annotated human instances was collected across diverse terrains in Jordan in collaboration with the Jordan Design and Development Bureau (JODDB). After preprocessing and mosaic augmentation, the dataset expanded to nearly 6000 training samples, enabling robust model fine-tuning. YOLOv8, initialized with VisDrone weights, achieved 97.0% precision, 97.6% recall, and 98.4% mAP@0.50. A multi-UAV routing algorithm based on a lawnmower pattern ensured 100% coverage of a 17.6 km2 pilot area using 16 UAVs with balanced mission durations. The results demonstrate that combining UAV clusters with AI-based detection significantly enhances scalability, coverage efficiency, and recall, reducing the risk of life-critical false negatives. While the system shows strong potential, challenges remain regarding communication constraints, latency, and environmental robustness. Overall, this work provides a validated framework for AI-supported UAV SAR operations and offers a foundation adaptable to broader disaster-response scenarios worldwide.​