In this letter, we propose a deep-unfolding-based framework (DUNet) to maximize the secrecy rate in reconfigurable intelligent surface (RIS) empowered multi-user wireless networks. To tailor DUNet, first we relax the problem, decouple it into beamforming and phase shift subproblems, and propose an alternative optimization (AO) based solution for the relaxed problem. Second, we apply Karush-Kuhn-Tucker (KKT) conditions to obtain a closed-form solutions for the beamforming and the phase shift. Using deep-unfolding mechanism, we transform the closed-form solutions into a deep learning model (i.e., DUNet) that achieves a comparable performance to that of AO in terms of accuracy and about 25.6 times faster.