In this talk, we present an offline-online strategy based on the Localized Orthogonal Decomposition (LOD) method for elliptic multiscale problems with randomly perturbed diffusion coefficient. The offline phase pre-computes entries to global LOD stiffness matrices on a single reference element for a selection of defect configurations. Given a sample of the perturbed diffusion the corresponding LOD stiffness matrix is then approximated by taking linear combinations of the pre-computed entries, in the online phase. We discuss error estimates of this approach and illustrate its performance with several numerical experiments.