A training process that teaches a model to refuse harmful requests and avoid generating unsafe content by reinforcing safer behaviors.