Drift control is the practice of keeping the governing authority stable over time by pinning customer-defined policies that do not change when underlying models are updated or retrained, and by treating material change as a trigger for review.
Model updates cause AI outputs to shift unpredictably. A response that was accurate and on-policy last month may become inconsistent or non-compliant after a provider updates their model. Organizations that depend on stable AI behavior need a way to see what changed and what authority still applies.
Drift control keeps the governing policy stable across model versions. Customer-defined policies remain the source of truth, and each governed output continues to receive an explicit decision against the pinned policy version. A material model change remains a trigger for review rather than something governance absorbs silently.
Without drift control, organizations discover governance failures after the fact. With it, the applied policy stays explicit and reviewable records continue to reflect the policy version that was actually applied to each governed decision.
Monitoring shows you what happened. Drift control enforces what should happen, before outputs are delivered.
A/B testing compares output variants. Drift control enforces consistent governance regardless of output variation.
Retraining changes model behavior. Drift control maintains governance stability independent of model changes.