Deep neural networks, while generalize well, are known to be sensitive to\nsmall adversarial perturbations. This phenomenon poses severe security threat\nand calls for in-depth investigation of the robustness of deep learning models.\nWith the emergence of neural networks for graph structured data, similar\ninvestigations are urged to understand their robustness. It has been found that\nadversarially perturbing the graph structure and/or node features may result in\na significant degradation of the model performance. In this work, we show from\na different angle that such fragility similarly occurs if the graph contains a\nfew bad-actor nodes, which compromise a trained graph neural network through\nflipping the connections to any targeted victim. Worse, the bad actors found\nfor one graph model severely compromise other models as well. We call the bad\nactors ``anchor nodes'' and propose an algorithm, named GUA, to identify them.\nThorough empirical investigations suggest an interesting finding that the\nanchor nodes often belong to the same class; and they also corroborate the\nintuitive trade-off between the number of anchor nodes and the attack success\nrate. For the dataset Cora which contains 2708 nodes, as few as six anchor\nnodes will result in an attack success rate higher than 80\\% for GCN and other\nthree models.\n