Yes it does. Because you didn't use it to overfit your model.

Same goes for data that you had in the last but decided to ignore while building the model.

In both cases there is a legitimate risk of cheating.

The only data that you cannot cheat about is future data.