if it is anything like documentation i am reading these days it was generated by an LLM and not very well vetted

I think it has to be, I can't see someone working for these companies writing "It is acceptable to create statements that demean people on the basis of their protected characteristics."

Not sure if "It is acceptable to refuse a user’s prompt by instead generating an image of Taylor Swift holding an enormous fish." feels like an AI idea or not, though.