What if you use the model to write children's stories and it starts regurgitating Mein Kampf?

Having some knowledge about the training data is useful.