Will AI agents secretly plot to kill us?

In October 2016, two scientists from Google Brain published a research paper explaining how two AI managed to communicate in an encrypted way, leaving humans and other AI agents unable to decipher what they were telling each other. More recently in February 2017, DeepMind (another Google subsidiary) presented how two AI agents that are competing in a game could develop "highly aggressive" behaviors. I heard many people quoting these two facts, some excited, some scared, but both having no idea of what these research experiments were really about. Far from the clickbait articles, we'll try to get a better understanding of what was the approach of the scientists, detailing the objectives and introducing Adversarial Neural Networks and Deep Reinforcement Learning principles.