Internet provides massive amounts of data. Millions of users publish contents and interact in a daily fashion. This opens the opportunity for developing and deploying a wide range of applications, including commercial tools and public surveillance services. By analising online data one can extract valuable cues about specific individuals, segments of users, geographic communities or the society as a whole. The eRisk campaigns have focused on stimulating research in extracting signs of different types of risk from Internet data. This chapter describes the main features of this experimental framework that, over the years, has brought together dozens of research teams to tackle different challenges related to early risk detection on the Internet.