How Do World Action Models Use Their Imagined Futures? Evidence from Mechanistic Interventions

World action models jointly generate predicted futures and robot actions for manipulation, but joint generation alone does not establish how the predicted future affects the action. Prior work shows that behavior depends on information at predicted-future token positions, but not how particular future content shapes the resulting action. We use two intervention stages to characterize that influence: how strongly and in what direction future content changes behavior, and which internal pathway carries the change. Stage 1 starts with two normal rollouts from the same simulator state. We insert the predicted future from one rollout (the donor) into the other (the recipient), hold all other inputs fixed, and measure whether the recomputed action moves toward the donor action, which the model never sees. Across 22 states and six RoboLab tasks, donor futures redirect Cosmos 3 actions and executed endpoints almost completely toward the donor rollout. Cosmos Policy shows a smaller effect at the first decision state of all ten LIBERO-Long tasks. Matched controls rule out generic perturbation and replacement as complete explanations. Content interventions show that Cosmos Policy responds primarily to camera-visible robot motion, whereas isolated robot or object pixels do not reproduce Cosmos 3's full predicted-future effect. In Stage 2, we keep the donor's predicted future inserted but restore the recipient's attention keys and values at predicted-future token positions. This removes 83–88% of Cosmos 3 steering in every tested state, identifying a pathway that carries most of the predicted future's causal influence.