To determine whether an AI model’s outputs areeffectively controlled for bias, an auditor needsempirical performance evidenceacross demographic groups.Similar accuracy ranges across groups (A)demonstrate that the model performs equitably and does not disproportionately advantage or disadvantage specific populations. This aligns with AAIA’s emphasis onfairness metrics, such as disparate error rates, equal opportunity, and demographic parity analyses.

Option B (fairness definition) is important for governance but does not prove fairness in practice. Option C may introduce historical human bias rather than mitigate it. Option D (transparency of development) supports explainability but does not validate fairness outcomes. Actualperformance parity across demographic groupsis the strongest evidence of bias control.