This paper delves into the application of Deep Reinforcement Learning (DRL) for Reconfigurable Intelligent Surfaces (RISs) to enhance wireless networks capabilities. RIS uses beamforming to reflect signals and is instrumental in improving network efficiency and service quality in B5G and 6G networks. Although DRL provides real-time adaptability, it also introduces security risks due to the lack of explainability in deep learning models. Our current research focuses on developing a simulation environment to rigorously test the robustness of DRL models against attacks such as eavesdropping. By analyzing these vulnerabilities, we aim to develop more resilient DRL models and effective mitigation strategies. This work is foundational for future research on the security of DRL-driven RIS, paving the way for more capable, secure, and robust communication networks.