That assumes they do so in a really stupid and straightforward way. LLMs already exist to "AI-wash" copyrighted material in ways that technically don't violate copyright. I'm pretty sure someone will find a way to create a dodgy shell company around a foreign B2B service that reycles this data for them in a way that is technically legal to use.

"Feed personal data into this service and it'll spit out a risk assessment based on a model built on 6.9M historical health data sets."