Abstract: With the booming development of the low-altitude economy, unmanned aerial vehicles (UAV) equipped with reconfigurable intelligent surfaces (RIS) are expected to become a key technology for constructing air-ground integrated secure communications. This paper investigates physical layer security transmission in a scenario where multiple ground devices transmit uplink data to a base station via a UAV-RIS relay in the presence of an eavesdropper. Specifically, the average secrecy sum rate is maximized by jointly optimizing the transmit power of ground devices, the phase shift matrix of the RIS, the two-dimensional (2D) flight trajectory of the UAV, and the receive beamforming of the base station, subject to the quality of service (QoS) requirements and UAV mobility constraints. To address the non-convexity and high coupling of variables in the formulated problem, an alternating iterative optimization algorithm based on block coordinate descent (BCD) is proposed. In particular, the successive convex approximation (SCA) is employed to handle the difference-of-convex (DC) structure in power allocation; the manifold optimization algorithm is adopted to tackle the unit-modulus constraints of the RIS phase shifts; the UAV trajectory optimization is transformed into a convex subproblem by incorporating the trust-region-based SCA; and a closed-form solution for the base station receive beamforming is derived based on the generalized Rayleigh quotient (GRQ) criterion. Simulation results demonstrate that the proposed scheme improves the average secrecy sum rate by 35%, 57%, and 73.5% compared to the fixed trajectory, random phase shift, and without RIS baseline schemes, respectively. This validates the effectiveness of the UAV-RIS cooperative architecture and the joint optimization strategy in enhancing the performance of low-altitude secure communications.