By exploiting idle time on volunteer's machines, desktop grids provide a way to execute large sets of computational tasks while requiring negligible maintenance and extra cost. Although, it constitutes a  valuable resource for low-budget projects, relying on external resources can be hazardous and ensuring the validity of the obtained results is mandatory. We will present a taxonomy of existing and possible risks by considering several aspects of workers faultiness (maliciousness, collusion, dynamicity). Then, we propose a general framework for scheduling tasks in order to obtain reliable results. Our final solution contains mechanisms for the scheduling and answer selection processes which use both a trust-based system that characterizes workers previous behaviors. Joint work with Louis-Claude Canon and Jon Weissman.