This study presents a comprehensive analysis of the critical and flow stress surfaces of a UO2 single crystal, computed using a newly developed machine learning interatomic potential (MLIP). The MLIP is based on the Spectral Neighbour Analysis Potential (SNAP) framework, and was built to accurately reproduce mechanical and defective properties of UO2. More specifically, it is based on an ab initio setup accounting for strong electronic correlations, relativistic effects and magnetic ordering, whose combination enables to predict isotropic elastic properties and a volume very close to experimental data. Following a validation stage with comparisons to experimental temperature trends, our potential is used to perform large-scale critical and flow stress surface calculations. Our results display an important directional dependence of the flow stress surface, and highlight the influence of temperature and strain rate on the mechanical behaviour of UO2. Key deformation mechanisms, such as dislocation nucleation and crack propagation, were identified, providing valuable insights into the material’’s response under various loading conditions. The effect of irradiation is also investigated, by comparing the obtained flow stress surfaces from pristine and defective UO2 samples. This work underscores the potential of machine learning techniques in advancing the understanding of nuclear fuel material mechanics.