Neutrality != necessarily accurate or useful. And the most neutral thing to say is nothing.

And most LLMs probably have Wikipedia as a significant part of their training corpus, so there is a big ouroboros issue too.

AI is itself biased because the training data is.