Jailbreaking as a Reward Misspecification Problem

Publication
In International Conference on Learning Representations