Recent research has revealed that the output of Deep Neural Networks (DNN) can be easily altered by adding relatively small perturbations to the input vector. In this article, we analyse an attack in an extremely limited scenario where only one pixel can be modified.