seems reasonable if there's ANY PII at all in the training set then the weights and any output should be treated as PII too of course the only way to verify t

seems reasonable if there's ANY PII at all in the training set then the weights and any output should be treated as PII too of course the only way to verify that a model doesn't contain PII is to provide the entire training set and the process for turning it into a model and opening up this data would completely destroy any economic benefit that could be gained from exploitation of the model which in turn completely destroys any incentive to create it in the first place what a pity