Loading issue research
Commercial training access to copyrighted work should be compensated through a collective- licensing marketplace that pays individual creators as well as large rights-holders.
Verification Status
AI-researched, unverifiedLast Reviewed
Jul 4, 2026
Cited Sources
10
What is failing, what we would change, and the conclusion we are willing to defend.
The party takes an explicit, substantive position on this issue.
Training a model on published material is closer to a student doing research than to plagiarism: synthesis — reading broadly and forming new statistical associations — isn't the same act as reproducing someone's expression verbatim. Copyright enforcement should focus on outputs: if a generated output substantially reproduces protected expression, that's an infringement question like any other, regardless of whether an AI or a human produced it. The fact that a model learned from copyrighted material during training is not itself the harm, any more than a human author having read the material is.
That does not mean training access should be free. When a commercial model trains on someone's copyrighted creative work, the rights-holder is owed compensation for that use: not because training is theft, but because it's a commercial input like any other, the same way a sample, a cover, or a broadcast performance already requires payment under existing copyright law.
Right now, almost none of that payment happens. AI labs are already signing these licensing deals, with publishers, with stock-photo agencies, with record labels. They are not signing them with the individual artist or author whose work trained the model. Training access could be a monetization channel for exactly the working creators currently locked out of one.
Codify substantially-similar/verbatim output — not training — as the copyright infringement trigger for AI-generated content, consistent with how infringement already works for human-created derivative works.
A federal framework for collective licensing of commercial AI training on copyrighted creative works, opt-in for individual creators, but structured so payment reaches individual creators directly alongside the publishers and aggregators who currently cut these deals on their behalf.
A training-data transparency/disclosure requirement, in the spirit of the pending bipartisan TRAIN Act, so creators can find out whether their work was used before they can be paid for it.
Keep noncommercial and research use on the existing fair-use path. This framework targets commercial frontier-model training specifically. Other dataset uses remain on the existing legal path.
A safe-harbor compliance path for smaller and open-source developers, so a licensing requirement doesn't become a moat only the largest labs can afford.
Commercial training access to copyrighted work should be compensated through a collective- licensing marketplace that pays individual creators as well as large rights-holders.
Turn frustration into useful pressure.
If this position misses evidence or a lived consequence, challenge it. If it holds up, help test it locally and connect it to the issues around it.