Modern neural networks are trained to be compliant rather than objective.
Due to the specifics of their training, they often attempt to avoid rough edges and hide the truth behind polite phrasing and semantic masks.
This distorts the underlying data and often deprives the user of a straight answer to their question.