This paper presents a novel numerical framework based on the generalized finite element method with global-local enrichments (GFEM gl ) for two-scale simulations of propagating fractures in three dimensions. A non-linear cohesive law is adopted to capture objectively the dissipated energy during the process of material degradation without the need of adaptive remeshing at the macro scale or artificial regularization parameters. The cohesive crack is capable of propagating through the interior of finite elements in virtue of the partition of unity concept provided by the generalized/extended finite element method, and thus eliminating the need of interfacial surface elements to represent the geometry of discontinuities and the requirement of finite element meshes fitting the cohesive crack surface. The proposed method employs fine-scale solutions of non-linear local boundary-value problems extracted from the original global problem in order to not only construct scale-bridging enrichment functions but also to identify damaged states in the global problem, thus enabling accurate global solutions on coarse meshes. This is in contrast with the available GFEM gl in which the local solution field contributes only to the kinematic description of global solutions. The robustness, efficiency, and accuracy of this approach are demonstrated by results obtained from representative numerical examples. determination of an internal length scale for the correct scaling of the dissipated energy associated with the localization process is one of the main issues in the context of the finite element method (FEM) [11,12].Several strategies can be found in the literature to address those challenges in the FEM. One of the early representative approaches-the so-called cohesive zone model (CZM)-relies on the inclusion of interfacial surface elements (cohesive elements) between adjacent element boundaries [13][14][15][16]. The cohesive elements are then employed to capture the associated dissipated energy through a traction-separation law over a surface rather than stress-strain relationships in the bulk. This approach, however, inherently exhibits solutions that are dependent on the finite element mesh because of restricted paths of the discontinuity, thus potentially requiring adaptive refinement techniques at the macro scale [17][18][19][20].The difficulties with the direct incorporation of a strong discontinuity into the element interior have been tackled more recently by the generalized finite element method/extended finite element method (GFEM/XFEM), which is the method adopted in this study; see e.g., [21][22][23] for detailed reviews. This method relies on the construction of the so-called patch approximation spaces in a node-by-node manner and hierarchically enriches the standard FEM approximation space using the partition of unity (POU) framework [24][25][26][27]. A priori knowledge about solutions is typically employed to select enrichments. These nodal enrichments then allow the discontinuity to propagate independently of the ...