Of course they test for accuracy. How do you think the realistic images end up looking realistic at all? It's not the only goal--it is generative AI, after all, and people mostly use it to create scenes that didn't actually happen--but a high degree of accuracy is a plus.

I don't see why you think they would even think to test that kind of accuracy, though. Feeling virtuous has a funny way of blinding you to potential negative consequences of your virtue-affirming actions. So you do your virtuous thing, get the right answer where you know you care, and done!

I just don't see how this isn't the most parsimonious model, as opposed to that they deviously trained the model to intentionally impose their untruthful view of the historical world on everyone and then were baffled by the reaction.