How do autoencoders detect anomalies?

Anomalies are detected using high reconstruction error.

Autoencoders are trained on normal data. When anomalous data is passed, reconstruction error increases because the model has not learned those patterns.