An approach is described for instructing a player character to automatically take cover in a region of a simulated environment without receiving express commands from a user. The approach determines whether the player character should take cover based on one or more factors. Such factors may include: the presence of an enemy-type character in the region; the presence of a nearby object in the region that can be used to seek cover; and a determination whether the player character's behavior is consistent with a decision to seek cover.