Observable indicators in an AI model's reasoning that reveal what action or goal it is planning to pursue.