An AI system governance review reveals that an AI model produces accurate predictions but uses features that encode protected characteristics through non-obvious proxies. What type of control should be implemented?

A is correct because feature importance analysis identifies which features drive predictions, and ablation testing (removing or neutralizing specific features) measures how much removing proxy features affects outcomes for protected groups. This is an evidence-based approach to diagnosing and quantifying discriminatory impact before deciding on the appropriate remedy. Discontinuing the model (B) is disproportionate without evidence that alternatives are better. Post-processing correction (C) may mask root-cause bias. Removing all correlated features (D) could eliminate legitimate predictive features.