Based on an emerging need for automated AI generation, we present a machine learning approach to generate behavior trees controlling NPCs in a “Capture the Flag” game. After discussing the game’s mechanics and rule set, we present the implemented logic and how trees are generated. Subsequently, teams of agents controlled by generated trees are matched up against each other, allowing underlying trees to be refined by learning from victorious opponents. Following three program executions, featuring 1600, 8000 and 16000 matches, highest scoring trees are presented and discussed in this paper.