Understanding the neural mechanisms responsible for human social interactions is difficult, since the brain activities of two or more individuals have to be examined simultaneously and correlated with the observed social patterns. We introduce the concept of hyper–brain network, a connectivity pattern representing at once the communication among the cortical regions of a single brain as well as the relations among the areas of two distinct brains. Graph analysis of hyper–brain networks constructed from the EEG scanning of 26 couples of individuals playing the Iterated Prisoner's Dilemma reveals the possibility to predict non-cooperative interactions during the decision-making phase. The hyper-brain networks of two-defector couples have significantly less inter–brain links and overall higher modularity - i.e. the tendency to form two separate subgraphs - than couples playing cooperative or tit-for-tat strategies. The intention to defect can be "read" in humans by evaluating the communication changes in the hyper-brain network.