Inference control

The discipline of keeping institutional meaning intact as it passes through AI systems, by placing a closed control loop around the model that compares outputs to an authorized reference and corrects drift.