We are trying to see if neural networks can learn to use secret keys to protect communications from other neural networks. Specifically, we are focusing on ensuring the confidentiality in a network with multiple agents and we specify these properties in the form of an adversary. Thus, we are trying to limit the information gleaned by Eve from a communication between Bob and Alice by eavesdropping on them. There are no specific cryptographic algorithms being assigned to these networks, rather we train them end to end adversarially.