When training data validation is inconsistent, the most severe risk is that the AI model may learn from incorrect, incomplete, biased, or corrupted data. This directly leads to a degradation of system reliability (option C), which manifests as inaccurate predictions, higher error rates, bias, or unstable behavior.
AAIA emphasizes that data validation prior to retraining is one of the most important controls because model behavior is fully dependent on training data integrity. If the quality and correctness of the data cannot be guaranteed, the resulting model outputs become unreliable, which can undermine compliance, operational decisions, and user trust.
Option A is less critical because increased complexity is not the core risk. Option B is important but secondary; documentation issues do not inherently degrade model reliability. Option D is an efficiency issue, not a risk to output integrity.
Therefore, compromised reliability due to poor-quality training data is the most significant risk.
[References:, AAIA Domain 2: Data Management Specific to AI (data validation, verification, data quality)., AAIA Domain 1: Governance and Risk Controls for AI., , ]
