During the first month when the company monitors the model for bias, it is most...

Theinitial deployment phaseof an AI model is critical forpost-deployment monitoring. When tracking forbias, the most important task is tocontinue disparity testingto determine whether outputs differ across protected groups.

From theAI Governance in Practice Report2025:

“Performance monitoring protocols… should include mechanisms to assess and measure disparities in outcomes across different demographic groups.” (p. 12)

“Bias may not be evident during pre-deployment testing but can emerge in real-world use.” (p. 41)

B. Awareness trainingis helpful, but not a technical bias mitigation activity.

C. Analyzing training datais apre-deploymenttask.

D. Documenting human decisionsmay support auditability but doesn’t detect bias in AI outputs.