Normalizing Test Data Correctly

Spot why the test set is normalized with the wrong statistics.

X_train = (X_train - X_train.mean()) / X_train.std()
X_test = (X_test - X_test.mean()) / X_test.std()

What is the bug in this normalization?