Could be they try to block explicit politics in general, for "broader audience, pretense of neutrality" reasons. Or it may be they don't want it providing misinformation on such sensitive topics; language models can easily "hallucinate" (make stuff up with a confidently wrong attitude, either because they don't know or because of a slight misstep in token probability). Also possible they are trying to catch certain kinds of things that aren't that and it hit a false positive on your attempt. Hard to say for sure with these models, without knowing their stated intentions (if they have any publicly stated to begin with).