Internet polls are becoming nowadays more and more important as they are being used on a large scale. Their security aspects are probably the most significant and hardest issue to be solved when we consider web surveys. In this paper we present an innovative solution which considerably increases the security of Internet polls and the reliability of their results by discriminating between human-cast and machine-cast votes. We define the basic idea of the method used to achieve our goal, its security properties and the testing results on a working prototype. In order to better test the performance and the effective robustness of the solution from the security point of view, we also propose a public challenge.