Governance

AI Change Management

Controlled processes for modifying AI systems while maintaining compliance and performance.

Definition

AI Change Management encompasses processes for safely modifying AI systems in production. AI changes present unique risks because model updates can have unpredictable effects on behavior.

Key AI change management considerations:

  • Model retraining: Updates to model weights based on new data
  • Feature changes: Adding, removing, or modifying input features
  • Threshold adjustments: Changing decision boundaries or confidence thresholds
  • Infrastructure changes: Updates to serving infrastructure affecting AI behavior
  • Vendor updates: Third-party AI API or model version changes

Best practices include: staged rollouts, A/B testing, rollback capabilities, impact assessment before changes, documentation of all changes, and re-validation after significant updates.

Sources

  • NIST AI RMF
  • ITIL Standards