<p>This paper investigates the secure consensus problem of continuous-time linear multi-agent systems with unknown system dynamics using a value iteration reinforcement learning approach. First, an integral sliding mode function is designed to drive the system trajectories onto the sliding surface, thereby eliminating the influence of attacks. Subsequently, by constructing the error system and exploiting the topological relationships among agents, a min-max strategy is employed to transform the secure consensus problem into solving the game algebraic Riccati equations. Furthermore, considering system dynamics are often challenging to obtain or involve high costs in practical engineering, a value iteration reinforcement learning method is adopted to compute the optimal consensus controllers, which removes the dependence on accurate system models and initial admissible control policies. Finally, a numerical example is provided to verify the effectiveness of the proposed method.</p>