We propose an alternating direction of multipliers method (ADMM) based on the consensus scheme for solving regularized multi-task problems where the loss function is separable across tasks. We discuss the convergence of the algorithm and survey its statistical properties in the sparse group Lasso case. Through numerical experiments, we find that our proposed method can be particularly efficient in the number of proximal evaluations required to achieve convergence when compared to a more general method such as FISTA.