Ensuring the safe and decent transmission of information in photonic networks, which serve as the backbone for high-speed Internet connectivity to billions of users, is paramount. Though, optical fibers, being the communication medium, are susceptible to various issues, including hard failures like fiber cuts and mischievous strikes such as optical monitoring (fiber tapping). These anomalies can lead to network disruptions, causing significant economical and information losses, compromising the privacy of information, and gradually degrading network operations. Hence, there arises an urgent requirement for robust detection of anomaly, identification, and locating strategies to improve dependability and accessibility of optical networks. This study introduces a data-driven approach aiming at precise, swift detection, diagnosis, and localization of fiber anomalies, spanning from fiber cuts to optical eavesdropping attacks. Our methodology integrates an autoencoder-driven anomaly detection mechanism with a bidirectional gated recurrent unit algorithm enhanced with attention mechanisms. The autoencoder is deployed for detecting faults, while the bidirectional gated recurrent unit is employed for false identification and pinpointing the fault location once the autoencoder flags an anomaly. achieves a 96.86% F1 score in detecting any fiber fault or anomaly. Moreover, the attention-focused bidirectional gated recurrent unit algorithm achieves an average accuracy of 98.2% in pinpointing detected anomalies and exhibits on an average rms error of 0.19 m in localizing faults.