[ttriton]

I think it would probably just be ignored.

What might have teeth though, is a modification to copyright law that does two things:

1. Make it mandatory for all AI projects to release their training dataset to interested parties, who may be copyright holders.

2. Change the default rights so that any company that wishes to use copyrighted content to train their AI model has to get an explicit opt-in for this purpose from the copyright holder.

With steep fines issued as penalties.

That should sort it out. Might also decimate the nascent large-scale AI model industry, but maybe that's what it needs before it goes truly out of control?