AI Fair Use Ruling: Your Judgment Is Property | Elacity
The Third Circuit's AI fair use ruling protected Westlaw's editorial judgment and named a licensing market for training data. Enforcement took six years. Here is how to make the license the lock.
The First Appellate AI Fair Use Ruling Called Your Judgment Property. Collecting Took Six Years.
If you write summaries, tag datasets, annotate cases, or curate anything an AI can learn from, a federal appeals court just told you that work is worth protecting. It also showed you how long the protection takes to arrive: the company that took it was gone years before the verdict.
On September 29, 2026, the Third Circuit affirmed that ROSS Intelligence infringed Westlaw's headnotes by using them to train an AI legal research tool. It is widely described as the first federal appellate fair use ruling on AI training, and its reasoning matters to anyone whose expertise ends up inside a model.
What the AI Fair Use Ruling Actually Decided
The panel held the headnotes were original because editors chose which holdings mattered, synthesized the facts, and drafted summaries that stand alone. Selection and judgment, not raw facts, were the protected asset.
On fair use, the court found the copying commercial and barely transformative, and counted harm not only to Westlaw's research business but to a developing market for licensing headnotes as AI training data. A court treating training-data licensing as a real market to be harmed is the quiet headline.
Concede the limits. ROSS was a non-generative tool, and the opinion leaves the generative AI question unresolved. Chatbot makers will argue their models transform far more, and that fight continues in other courts.
The Failure of Enforcement by Lawsuit
Thomson Reuters sued in May 2020. By December that year, ROSS announced it would shut down because the litigation blocked its fundraising. The appellate win landed roughly six years after the copying it addressed.
Now scale that down to you. Thomson Reuters had in-house lawyers, a famous product, and years of patience. A freelance annotator, an independent researcher, or a small newsletter has none of those, and the doctrine protecting their judgment is identical.
Publishers with resources keep filing anyway. USA Today Co. sued OpenAI on October 8, alleging its articles were used for training without a license; the claims are unproven and the case has barely begun. Each new suit restates the pattern: the copy happens in seconds, the reckoning takes years.
We have argued before that when policy leans toward fair use, your only enforcement left is the key. This ruling leans the other way and still proves the same point: winning the argument is not the same as controlling the file.
The Paradigm Shift: Make the License the Lock
The Third Circuit recognised a licensing market. Markets need a place where terms are checked before delivery, not litigated after copying. That is the gap Elacity is built for: turning curated knowledge into property you hold, where access is the license.
For the full model of who holds the key, start with our explainer, Decentralized DRM, Explained. Here is how it applies to the exact thing this court protected.
1. Package the judgment, not just the file
With Elacity dDRM, a set of headnotes, annotations, or a labelled dataset becomes a Wealth Capsule: an encrypted, programmable on-chain good with your rights and royalty terms written into it. The curated work never leaves your hands in the clear; what travels is a sealed object and its terms.
2. The key checks the license every time
The key that unlocks a capsule is split across an owned 2-of-3 quorum of independent machines. No single operator, Elacity included, holds it, and each machine re-checks the buyer's on-chain rights before releasing its share. No valid license, no key; the check happens before use rather than in a courtroom afterward.
3. Used inside a sealed moment, then wiped
Content stays encrypted everywhere except a sealed sandbox at the moment of use, where the key exists in the clear only for that one operation before it is wiped. Whatever runs the job, a human reader or a training pipeline, gets the use it paid for, never a reusable key. Humans and AI pass through the same capability gate, with narrow, expiring permissions that fail closed when revoked.
The honest edge: this is trust-minimised, not trustless, since a colluding quorum could in principle reconstruct a key. Agent wallets and an autonomous approval loop for AI buyers are still being built; the primitive they rest on, a key an agent can use but never see, already exists. ElastOS, the open-source runtime underneath, runs fullest on Linux today.
Old Internet Versus Owned Knowledge
- Old internet: your curation is scraped, then you sue. Elacity: your curation is sealed, and access is the license.
- Old internet: proof of harm arrives years later. Elacity: rights are checked at every unlock.
- Old internet: only those who can fund litigation get protected. Elacity: the same gate works for a solo annotator and a publisher.
None of this replaces the law; it gives the law something to point at. If you want to go further, our guide on how to turn your data into capital walks through what makes a dataset worth packaging. More coverage of rulings and rules lives in Ecosystem & Governance.
As Elacity's founder, Sasha Mitchell, puts it: "The people who create the value should own it. That is the entire reason Elacity exists."
The court has now said your judgment is property. The next step is holding it like property. Follow Elacity on X for how we're building that.