Request the article
A forthcoming law-review-style article details the doctrinal basis for generation-moment licensing. Request early access.
Or email us at contact@yodal.aiThe case for a transaction layer
A generation-moment licensing architecture offers something litigation and lump-sum deals cannot: a price signal that tracks actual use. This page outlines the problem, the mechanism, and the legal scaffolding required to make it work.
01
The pricing problem
Generative AI has created the largest unpriced market in the history of creative work. Models train on billions of works; outputs draw on that training in ways neither the developer nor the rightsholder can quantify. The result is a market with no price signal.
Four responses have emerged. None clear the gap:
- Bilateral deals — lump sums unlinked to actual output. A bet on average use, not measured contribution.
- Linked referral — prices the door (a link to a publisher), not the use. Misaligned with the underlying value.
- Post-hoc detection — finds outputs that resemble training data. Resemblance is not contribution.
- Inference access — the auditor gains access to the model. The developer still holds the data. The trust problem remains.
What's missing is a price signal that tracks contribution at the moment of generation — a transaction layer that neither side controls.
02
The mechanism
Yodal's attribution engine runs at inference time. When a model generates an output, the engine searches a registry of works for likely contributors. Each candidate receives a contribution score. If a rightsholder's work likely contributed, and if the developer has opted in to that rightsholder's terms, the license executes automatically.
The architecture is designed to be neutral by construction:
- No models — Yodal does not train or deploy generative AI.
- No content — Yodal does not own or license creative works.
- No pricing — Rightsholders set their own terms. Yodal measures and executes.
The result is a market where both sides can see measured contribution and negotiate accordingly — without either side ceding data or control.
03
The legal landscape
Two forces are converging on the same architecture:
Litigation
More than 100 U.S. cases are in active litigation. Recent decisions suggest courts are receptive to market-harm arguments — but market harm is hard to prove when the market has no price signal. A transaction layer provides that signal.
Regulation
The EU AI Act requires training-content summaries. California's AI disclosure law takes effect in 2026. Federal proposals (CLEAR, TRAIN, NO FAKES) are pending. The regulatory direction is toward per-use accounting — exactly what a transaction layer provides.
A forthcoming law-review-style article by Yodal's founder details the doctrinal basis for generation-moment licensing. The attribution architecture and the legal mechanism were designed together.
04
The architecture
The trust problem that killed prior audit approaches: any method that requires training-data disclosure fails, because developers won't disclose and publishers can't verify without disclosure.
Yodal's architecture is designed to break this deadlock.
It attests to contribution without revealing what was trained. This is the property that makes generation-moment licensing viable.