They should have to make the weights public if they train on data scraped from the public internet. It’s be something like copyleft by default for AI training.
How do you track down the sheer number of them so efficiently? You could obviously dig through each one’s profile, but that wouldn’t work for hundreds or thousands.