Is Autoencoder only trained with normal samples for Anomaly Detection?

[TAsUjxnMIL]

Hello,

I have looked into "PyOD Tutorial - autoencoder.ipynb", as I understood the X_train dataframe contains normal and anomalous samples. This is the way I also would choose, but in other explanations/articles the model is fitted only with normal samples. So I am not sure which one is the correct one, or are both ways correct?

I would be grateful, if you could answer me this question.
Thank you!