Obviously overfitting is always a concern, but deep learning has (when done correctly) found ways to mitigate concerns about overfitting.

Dropout, for instance, or k-fold cross validation.