Crowdsourcing platforms dedicated to work, be it paid or voluntary, essentially consist in intermediating between tasks—sent by requesters—and workers. They are used by a growing number of individuals and organizations, for tasks that are more and more diverse, complex , and that require specific skills, availabilities, experiences, or even devices. On the one hand, these highly detailed task specifications and worker profiles enable high-quality task assignments. On the other hand, detailed worker profiles may disclose a large amount of personal information to the central platform (e.g., personal preferences, availabilities, wealth, occupations), jeopardizing the privacy of workers. In this paper, we propose a lightweight approach to protect workers privacy against the platform along the current crowdsourcing task assignment process. Our approach (1) satisfies differential privacy by building on the well-known randomized response technique, applied by each worker to perturb locally her profile before sending it to the platform, and (2) copes with the resulting perturbation by benefiting from a taxonomy defined on workers profiles. We describe the lightweight upgrades to be brought to the workers, to the platform, and to the requesters. We show formally that our approach satisfies differential privacy, and empirically, through experiments performed on various synthetic datasets, that it is a promising research track for coping with realistic cost and quality requirements.