This letter considers the secure communication in a reconfigurable intelligent surface (RIS) aided full duplex (FD) system. A FD base station (BS) serves an uplink (UL) user and a downlink (DL) user simultaneously over the same timefrequency dimension assisted by a RIS in the presence of an eavesdropper. In addition, the BS transmits artificial noise (AN) to interfere the eavesdropper's channel. We aim to maximize the weighted sum secrecy rate of UL and DL users by jointly optimizing the transmit beamforming, receive beamforming and AN covariance matrix at the BS, and passive beamforming at the RIS. To handle the non-convex problem, we decompose it into tractable subproblems and propose an efficient algorithm based on alternating optimization framework. Specifically, the receive beamforming is derived as a closed-form solution while other variables are obtained by using semidefinite relaxation (SDR) method and successive convex approximation (SCA) algorithm. Simulation results demonstrate the superior performance of our proposed scheme compared to other baseline schemes.