the "true believers" at OpenAI mostly don't care if bad images get generated. they are worried about "safety" as in the sci-fi Terminator scenario, not "safety" as in "avoid harming people with offensive or unpleasant images".

however they see controlling a cutting edge AI towards some goal to be an important thing to study, and they want to practice doing it, and to do that they need to pick some arbitrary goal.

the arbitrary goal becomes this type of "PG-rated-only" censorship, because it helps avoid bad press coverage, makes it easier to raise money, etc. but they don't sincerely care about it. some others in tech do sincerely care though.