You are concerned about what you perceive as post-facto editorializing. But I think that glazes over the fact that human bias and politics are already built into every AI learning model at the data-labeling phase. No AI model is ever really pure or unfiltered, they are fundamentally a reflection of how the developer views the world from the outset. I am not really bothered by any additional guardrails put on to make sure it errs on the side of caution when it comes to certain topics.
This idea that you should be able to use an AI model without any understanding of who built it is false. It's like reading the news. You know that certain publishers have their political perspectives, and you can read their perspectives while understanding their blind-spots and motivations and you can choose to believe them or look for other perspectives or have a nuanced understanding of the topic. The same is true for AI usage. Research the team that created it, read their ethics statements, and decide if that model is right for you. It's a literacy problem, your rights aren't being taken away because of someone's design choices.