Climate change has led to a significant increase in the frequency and intensity of landslides, debris flows, and other rapid mass-movement hazards, thereby escalating risks to critical infrastructure. As many protective systems are now nearing the end of their service life, the development of next-generation, resilient structural designs requires updated engineering guidelines based on advanced physics-based numerical modeling. Depth-averaged shallow-water models are commonly used for rapid hazard assessment and for predicting the runout and final deposition of the landslide. However, these models have limitations due to how they are formulated. By averaging flow properties over depth, they fail to capture the three-dimensional behaviour of the material. This results in a loss of detailed information needed to simulate soil–structure interaction and resolve impact forces on protective structures. More comprehensive three-dimensional continuum methods are needed to evaluate structural performance under extreme loading conditions. This study investigates the impact forces generated by rapid granular flows on protective structures, such as retaining walls and barrier nets. The Material Point Method (MPM) is chosen due to its capability to handle large deformations and complex flow–structure interactions. MPM, however, faces a challenge in the form of non-physical oscillations of the stresses. These oscillations come from well-known numerical artefacts in MPM, such as cell-crossing errors, small-cut instabilities, and mapping errors. This leads to highly oscillatory reaction forces, which impede convergence and compromise the reliability of coupled simulations. The influence of these artefacts is analysed, and several mitigation strategies are evaluated. The results improve the stability and accuracy of coupled MPM simulations, ultimately making the method more reliable for protective structures exposed to extreme natural hazards. All simulations and numerical developments are implemented within the Kratos Multiphysics.