Model change means any update that alters how a system behaves, makes decisions, or processes data. In security, this can include retraining a machine-learning model, changing features or thresholds, swapping a decision engine, or modifying a workflow that feeds data into automated controls. Even small changes can shift outputs, risk scores, alerts, or access decisions.
It matters because model changes can quietly invalidate assumptions that defenses, audits, and approvals depend on. Attackers may try to exploit a poorly reviewed change by poisoning training data, manipulating inputs, or pushing an unsafe configuration into production. Defenders reduce that risk with change control, approval records, testing, rollback plans, and post-change monitoring so they can prove the system still behaves as expected after the update.