High dimensional regression benefits from sparsity promoting regularizations. In such a context, screening rules leverage the known sparsity of the solution by ignoring some variables during (or even before) the optimization process, hence speeding up solvers. Such rules are said to be "safe" when they cannot wrongly discard features. In this talk, new safe rules for generalized linear models with sparsity enforcing regularization will be proposed. Our proposed GAP Safe (screening) rules can cope with any iterative solver and we illustrate their performance on coordinate descent, demonstrating interesting speed ups for learning problems. This is a joint work with E. Ndiaye, O. Fercoq and A. Gramfort