A regulator asks a bank to explain why its AI fraud-detection model blocked a specific customer's transaction. The data science team confirms the model is accurate overall, but they cannot identify which factors led to that particular decision or justify it to the customer. Which AI risk does this situation BEST illustrate?

Explainability is the ability to describe how an AI system reached a particular output in terms that people can understand and verify. Complex models can be accurate in aggregate while remaining opaque about individual decisions, which creates problems when regulators, auditors, or affected customers are entitled to an explanation. The scenario shows exactly that gap: overall accuracy is acceptable, but the team cannot identify the factors behind one decision or justify it. Hallucination refers to generative models producing fabricated content, not to a classifier whose decision cannot be explained; the block may even be correct. Bias describes systematic unfairness toward particular groups, and nothing in the scenario indicates a pattern of disparate treatment, only an inability to explain one outcome. An evasion attack involves an adversary crafting inputs so a model misclassifies them, and there is no adversary or misclassification described here. Exam caveat: explainability is a governance and accountability concern rather than an attack, and it often drives choices between simpler interpretable models and more complex ones. Operational check: implement explanation tooling that records the principal factors behind each decision, retain decision logs, provide a human review and appeal path for blocked transactions, and document model behavior for regulators.