AI-driven technical support systems rely on accurate, current knowledge to resolve user issues. Model drift causes the system to diverge from real-world conditions, producing inaccurate outputs that erode user trust, increase escalations, and potentially cause harm if incorrect technical guidance is followed.

Why A is Correct: According to ISACA AAIR validation guidance, inaccurate outputs from model drift represent the greatest risk in a technical support AI because they directly compromise the system ' s core function—providing correct technical guidance. Inaccurate outputs lead to unresolved issues, potential system damage from wrong instructions, and reputational harm. Unlike the other options, drift-driven inaccuracy affects every user interaction and cannot be remediated without model updates.

Why B is Correct Context: Infrequent training dataset updates are a contributing cause of model drift and are a serious concern, but they are an input factor rather than the manifest risk itself. The concern is the resulting inaccuracy.

Why C is Wrong: Encryption is a security control for data in storage and transit. While important for confidentiality, it does not affect the accuracy of AI outputs or the system ' s ability to provide correct technical guidance.

Why D is Wrong: Excessive manual sampling is a testing methodology concern that may reduce testing coverage efficiency. However, it represents a process inefficiency rather than a direct risk to output quality—the model ' s accuracy is the greater concern.