An AI system used for medical image analysis begins producing incorrect diagnoses six months after deployment despite unchanged inputs. Root cause analysis reveals the model's training data was collected from a different patient population. This scenario best illustrates which risk?

A is correct because covariate shift occurs when the statistical distribution of inputs in production differs from the training distribution, causing degraded performance even when the underlying relationship (label logic) has not changed. B is wrong because label noise refers to annotation errors, not population differences. C is wrong because a backdoor attack is a deliberate malicious implant. D is wrong because overfitting to validation data would manifest as poor initial performance, not degradation after months.