An issue with regulating the "politics" of LLMs is that, while LLMs have biases, they can be subtle. A lot of what looks like bias in testing models is really sycophancy. GPT-3.5 inferred your political beliefs (even from what football team you like!) & tried not to upset you.