I think there's still an ethical question of a business trying to scoop a researcher solely because the business learned of promising progress. It's not a very big leap to consider "multiple discovery" (a la calculus by Leibniz and Newton) and chance that the model that is good at math proofs might have enough ambient context to work faster from a similar knowledge base on the problem for which promising progress was discovered. I think that's not nice at the very least. Surely the above-board way to approach that is "hey researcher, we learned, indirectly, that you might have made significant progress on flashy research topic and would like the chance to throw our vast computational resources in to try to help work toward the conclusion of that effort." But, maybe that's asking too much?