Model validation
Evaluating whether a model performs as expected, including its reliability and its limitations.
Model validation is the set of activities that evaluate whether a model performs as expected, including an assessment of its reliability and its limitations. Under the US agencies' revised model risk guidance, the rigour of validation follows the model's approach, use and materiality, generally takes place before first use, and covers elements such as conceptual soundness and outcomes analysis. For an agent, validation extends to the whole system around the model: its tools, its limits and the checks that stop wrong actions.
Agent Minute explains this term on 26 October 2026.
Related terms
Model riskThe potential for adverse consequences from decisions based on model outputs, including misuse of a sound model.Effective challengeCritical, objective analysis of a model by experts with the independence and standing to force changes.Ongoing monitoringContinuing checks that a model or agent is still performing as intended as data, use and conditions change.Red-teamingStructured adversarial testing of an AI system to find flaws, harmful behaviours and vulnerabilities before attackers do.