AI & Copyright / Training Data Rights

The strongest good-faith objection: a collective-licensing clearinghouse, however well designed, still requires someone to decide how to value an individual creator's contribution to a model trained on an enormous corpus of works. And a critic could reasonably argue this is either practically unworkable at the scale required (already partly conceded elsewhere in this issue) or, if simplified enough to be administrable, will systematically underpay individual creators relative to what a fair per-work valuation would require. That's an argument for building the clearinghouse with per-creator payment floors and independent valuation audits from day one, not an argument against building it at all. The alternative already in place, where large rights-holders capture licensing deals and individual creators get nothing, is strictly worse than an imperfectly-calibrated clearinghouse that routes payment to people currently getting none.

AI companies bear new compensation costs, which could be passed through to consumers via pricing. Individual creators may still be underpaid relative to large rights-holders even under a clearinghouse model if its valuation formula favors scale or aggregation. Smaller and independent AI developers without major-label-scale licensing budgets could face a barrier to entry that better-resourced incumbents absorb more easily. Proposal 3's revenue- and compute-scaled safe harbor is built for exactly this: it exempts smaller and open-source developers from the clearinghouse cost structure entirely below the threshold, which is the deliberate design answer to this risk, not a residual one this position leaves unaddressed.