I think the concern you raise about data ownership is very real and increasingly so with the maturation of AI technology.
At the same time I want to push back a little against the notion of predicting the future based on the past. It is a tricky thing. Others, particularly Nassim Nicholas Taleb comes to mind, have laid out the involved problems in their works. I tackled the issue with regards to AI in this piece. As a heuristic I propose: The more complex a domain is (the more variables are or ought to be considered) the harder it gets to make predictions with any precision. Another, related problem is faux correlation that derives from large data sets. I think it’s necessary for AI developers to be aware in what environment they operate in.
That being said, the problem you hint at is indeed critical. Solutions are very much welcome!