See Also: Red Teaming, Interpretability, and Monitoring
Alignment assurance leans on three practices that get full treatment elsewhere in this book. Rather than duplicate that material, this section points to where each one lives.
Red Teaming
Structured adversarial testing — threat models, attack surfaces, adaptive probing, evidence capture, and retest loops — is covered in Unit 4: Adversarial AI, Chapter 4: Red Teaming. Those tools apply directly here: the target is unwanted or misaligned behavior across the deployed system.
Interpretability and Explainability
Methods for revealing learned features, circuits, and internal strategies are covered in Unit 1: Interpretable ML. Methods for explaining individual predictions and global model behavior are covered in Unit 2: Explainable Machine Learning. For alignment work, treat explanations as testable evidence: pair them with an intervention rather than accepting them as a story that merely fits the outcome.
Monitoring and Auditing
Deployment monitoring and audit practice — metrics, thresholds, ownership, and audit trails — are covered in Unit 6: Transparent AI, Chapter 2: Accountability, Monitoring and Auditability.
Come back to these three units whenever an alignment assurance question turns into "how do we actually test this" or "how do we know this is still true after deployment."