RL approaches that enforce safety constraints during training and deployment to prevent harmful actions.