I think approving the comment is also doing something, so if we won't allow it to do that too it will be useless. But I think you mean we shouldn't allow AI to do potentially harmful actions like removing/rejecting comments. But approving a bad comment also means people see it so it's a debate about which one being more harmful in practice. But I agree with you!