I don't see how it would cause legal issues, a language model isn't their representative, so anything it says doesn't affect the company.

I'm not a lawyer, but why do all the other big companies filter their LLMs then?

I really think that people would use it for bad, and people would at least try to take legal action against those creating/hosting the model.

Not that they would lose, but it seems like there would be legal annoyances to deal with at the very least if they don't filter their outputs.