Data leakage
Also: target leakage, label leakage, train-test contamination
Leakage happens when features encode the answer, when preprocessing is fit on the full dataset before splitting, or when the test set is used to make modeling decisions. The model looks great in evaluation and then fails in production.
Example
Standardizing features with the mean and variance of all 10,000 rows before the split lets test-set statistics leak into training; fitting the scaler on the training fold only fixes it.