Physical layer security (PLS) constitutes a crucial foundation for future trustworthy networks, particularly in scenarios involving low-resource devices, low-latency, and high-mobility applications. The effectiveness of PLS can be significantly enhanced through the integration of jamming and anomaly detection mechanisms that exploit the inherent randomness of the wireless channel and leverage machine learning (ML) techniques. Our approach capitalizes on 5G new radio (NR) sounding reference signal (SRS) measurements to facilitate accurate channel estimation and its corresponding SNR estimation, thereby enabling the identification of anomalous behavior that may be indicative of a potential attack. For the learning-aided detection, various features are extracted from the estimated channel state information (CSI), particularly the channel impulse response (CIR) of the 5G NR SRS. Through the training of an ML classification model on a comprehensive dataset of SRS measurements collected in our laboratory, we achieved a detection accuracy above 90%, investigated feature importance, and demonstrated the viability of this approach in bolstering PLS in future wireless networks.