In recent years, seismic safety requirements for nuclear power plants have become increasingly stringent, highlighting the growing need for computationally robust structural integrity assessment frameworks under extreme loading conditions. However, crack propagation criteria applicable to extremely low cycle fatigue (ELCF) induced by severe earthquakes have not yet been fully established within the context of nonlinear finite element analysis[1]. As a result, structural components subjected to excessive cyclic loading accompanied by complex multiaxial stress states are often evaluated using overly conservative design and maintenance strategies, leading to increased lifecycle costs. In a previous study, a crack propagation criterion was proposed based on fracture experiments using 1T compact tension (1TCT) specimens fabricated from austenitic stainless steel SUS316, and was subsequently implemented in a finite element–based crack propagation framework. The criterion employed the increment of equivalent plastic strain and stress triaxiality, which are key parameters governing ductile fracture, evaluated at the mid-thickness integration point of the specimen. Although a correction function was introduced to account for the reduction in stress triaxiality near the free surface, the proposed criterion failed to accurately reproduce crack propagation behavior in the three-dimensional transition region, where the through-thickness gradients of stress and plastic strain become pronounced[2]. In the present study, a new computational crack propagation criterion based on cumulative equivalent plastic strain is proposed to enhance predictive accuracy under ELCF conditions involving repeated cyclic loading and large-scale plastic deformation. Three-dimensional finite element simulations demonstrate that the proposed criterion successfully reproduces surface-near crack propagation behavior, which has been difficult to predict using conventional criteria based on local strain increments. These results suggest that the proposed framework provides a numerically consistent and physically interpretable basis for evaluating crack propagation under ELCF conditions, and offers strong potential for high-fidelity integrity assessment of safety-critical structures subjected to extreme events.