With the recent advances in voice synthesis such as WaveNet, AI-synthesized fake voices are indistinguishable to human ears and widely applied to producing realistic and natural DeepFakes which are real threats to everyone. However, effective and robust detectors for synthesized fake voices are still in their infancy and are not ready to fully tackle this emerging threat. In this paper, we devise a novel approach, named DeepSonar, based on monitoring neuron behaviors of speaker recognition (SR) system, a deep neural network (DNN), to discern AI-synthesized fake voices. Layer-wise neuron behaviors provide an important insight to meticulously catch the differences among inputs, which are widely employed for building safety, robust and interpretable DNNs. In this work, we leverage the power of layer-wise neuron activation patterns with a conjecture that they can capture the subtle differences between real and AI-synthesized fake voices and provide a cleaner signal to classifiers than raw inputs. Experiments are conducted in three datasets (including commercial products from Google, Baidu, etc) containing both English and Chinese languages to corroborate the high detection rates (98.1% average accuracy) and low false alarm rates (0.02 equal error rate) of DeepSonar in discerning fake voices. Furthermore, extensive experiment results show its robustness against manipulation attacks (e.g., voice conversion and additive real-world noises). Our work also poses a new insight into adopting neuron behaviors for effective and robust AI aided multimedia fakes forensics as an inside-out approach instead of being motivated and swayed by various artifacts introduced in fakes.