Adversarial attacks pose severe threats to deep learning systems, especially in safety-critical domains such as video surveillance, robotics, and autonomous driving. Among these attacks, adversarial patches have gained attention due to their localized, printable nature and effectiveness in deceiving object detectors. In this paper, we investigate the generation of adversarial patches using generative adversarial networks (GANs), evaluating two different generator architectures (DCGAN and self-attention) against multiple YOLO variants (v3, v5, v8, and v11). We demonstrate that these patches can reduce the mean Average Precision (mAP) of YOLO detectors by over 20 points and that they exhibit partial transferability across different YOLO versions. Furthermore, we discuss theoretical underpinnings of adversarial patch formation and possible defenses, such as adversarial training and anomaly-based detection, to mitigate these threats. Our findings underscore the vulnerability of even the latest YOLO architectures, highlighting the pressing need for robust security measures in critical vision applications.