I think the problem is that “transformation” has never been clearly defined in copyright law, and that ambiguity is exactly what AI companies lean on in their defense.

At a human scale, those boundaries get clarified through litigation on a case by case basis once an infringement becomes large or obvious enough. But there has always been a gray area around when a derivative work crosses the line into infringement.

And AI didn’t create that ambiguity, it just pushed it to a more extreme scale