DAMAGE: Detecting Adversarially Modified AI Generated Text