It had memorized the information in such a way that it could be recovered by any user—even without access to the original dataset. This problem is known as unintentional memorization and can be prevented with appropriate security measures—or by keeping sensitive data out of a neural network’s training dataset in the first place.
