Are AIs more likely to pursue on-episode or beyond-episode reward?

RL would encourage on-episode reward seeking, but beyond-episode reward seekers may learn to goal-guard.