> Using someone’s handwriting as training data for OCR, as part of a training dataset significantly large enough to be used as training data in the first place,

> Using someone’s handwriting as training data for OCR, as part of a training dataset significantly large enough to be used as training data in the first place, does not pose a material additional privacy concern. USPS literally processes people's private addresses. I'm amazed how you can say "Does not pose privacy concern" with a straight face when the entirety of the past decade has shown us that if there's private data, companies will abuse it, sell it, leak it etc.