While I'm at it, add a requirement that the output include a link to every piece of content the model was trained on for which the company doesn't have a contract/license/release from its creator.
Solves the uncredited, used-without-permission problem.
@cazabon It's not clear that even including the links will do any good. Hardly anyone will bother clicking them. The result is the GAI is a taker of data and gives little or nothing in return, not even the clicks and views that traditional SERPs generate.