The cencoring isn't even built in, that's why you see it give a response for a few seconds and then being cut off by probably a secondary model whose job is to censor the output.