Adversarial ML Attack that Secretly Gives a Language Model a Point of View