They try their best to keep their business as profitable as possible. This often means they just don't want to analyse content in a sufficient way, although it should be possible to do it in an automated way. It's better to take down 10 videos with the word pedophilia, even if 9 of them talk about how to make your kid safe, than to spend computing power on a more detailed analysis. Why bother, those videos won't sell ads anyway. It's one of the reason that the whole "I'm a private company na na na na na na ha ha ha ha ha ha, you can't regulate me!" argument is so weak.

Not every problem can be solved with automation, because content adapts. If there will be people involved, there will be biases and scandals. Not that I trust people designing algorithms to be some kind of saints.

We know now that people are more likely to discriminate basing on political orientation than race. Basically all the people that would be racists 100 years ago now took a different angle to be able to behave in the same manner. I do not trust moderators and overlords, because they seem to be all from one political orientation. Being a black person 100 years ago, would you trust an all-white private company to handle all your information in an equal way, even knowing that some of the data exposes illegal practices of the said company? It's absurd to think that anyone under those circumstances would be 100% clean and it's just plain insane to thinnk we can hold hundrets of thousands of people at the same time to thius standard.