I, at this point in time, don't believe any vetting of the training materials was done, because they needed too much of it to make their ML model work. They did

I, at this point in time, don't believe any vetting of the training materials was done, because they needed too much of it to make their ML model work. They didn't check for licenses. They didn't check for content. They didn't check for correctness. At this point, it's cheaper for them to manually alter the output than it is to retrain on a more appropriate data set.