Within the ISACA Advanced in AI Risk framework, life-cycle controls should protect data quality, model design, testing, validation, monitoring, change management, and secure retirement of AI systems. Cleansing and normalization reduce inconsistent, erroneous, or statistically distorted training data that can produce biased or inaccurate model behavior. Their primary value is improving trustworthy inputs rather than lowering compute cost or satisfying privacy law by themselves. This makes option C, Addressing statistical distortions in training data that could cause model bias, the strongest answer. The other choices describe narrower technical, operational, performance, or administrative considerations and do not address the primary risk-management objective in the scenario as directly. A risk practitioner should select the response that most effectively reduces the stated exposure while preserving appropriate oversight, traceability, and alignment with organizational risk tolerance and business requirements.