Comment by lazide

If both models pull from the same training data, wouldn’t it likely just be enough to poison the common training data?