These responses closely mirrored examples of the false claim from pro-China sources, which alleged that Taiwan’s Democratic Progressive Party (DPP) was suppressing opposition voters by deliberately withholding voter notifications.
There's two things at play here. First, all models being released these days have safety built into the training. In the West, we might focus on preventing people from harming others or hacking, and in China, they're preventing people from getting politically supportive of China. But in a way, we are all "exporting" our propaganda.
Second, as called out in the article, these responses are clearly based on the training data. That is where the misinformation starts, and you can't "fix" the problem without first fixing that data.