Motivated by modeling borrowing and lending between banks, we start by illustrating systemic risk with a toy model of diffusions processes coupled through their drifts. We then show that such a simplistic model is in fact a Nash equilibrium of a Linear-Quadratic differential game. In order to take into account clearing debt obligations a delay component is introduced in the game and we describe how it affects liquidity and systemic risk characterized by a large number of defaults. We use this toy model to discuss Mean Filed Games and Mean Field Control problems as the number of players becomes large, as well as numerical methods based on Deep Learning techniques. This is a joint work with R. Carmona, M. Mousavi, L.H. Sun, and Z. Zhang.