AI Training Fair Use: Enforce It at the Key | Elacity
Washington told a court that training AI on your work is fair use and pointed creators to Congress. Copyright never stopped a copy. The only enforcement left is the key.
The DOJ Says AI Training Is Fair Use. Your Only Enforcement Left Is the Key.
A federal court is now weighing an argument the United States filed in your name: that copying your work to train an AI model is fair use, and that you are owed nothing for it. The same filing tells you where to take your objection. Not to the court. To Congress.
On September 1, the Justice Department urged a New York federal court to hold that training a large language model on copyrighted work is fair use. The brief calls the copying extraordinarily transformative and warns that rules making American AI harder to build threaten national security and hand an edge to foreign adversaries. For creators, the practical message was blunt: if you want to be paid, your remedy is Congress, not the courtroom.
What the Fair Use Filing Quietly Settles
A Statement of Interest is not a ruling. It is not binding on the judge, who can give it as much or as little weight as he chooses. And the fair use argument has real force. A model that absorbs statistical patterns is not a bootleg copy sold from a competitor's shelf, and treating every training pass as theft would freeze work that is genuinely new.
Concede all of that, and the direction still holds. The most powerful party in the room has told the court that feeding your work to a model is lawful, even as the same government negotiates a stake in the company it is defending. When the referee has a wager on the game, waiting for the whistle is not a strategy.
Enforcement Was Always the Real Question
Copyright never stopped a copy. It let you sue after one was made. That was enough when copying was slow, visible, and expensive. The remedy assumes a world where you can find the copy, name the copier, and afford the fight. A training run that ingests millions of works in a weekend erases all three, and leaves no seam to follow.
You can see the gap in every settlement. Anthropic agreed to pay authors about 1.5 billion dollars over books pulled from shadow libraries. The sum was historic. The files were still taken first, and the money arrived years later, by lawsuit, on terms set by the taker. We made the same case about licensing in Renting the Tollbooth: a payment you did not set and cannot enforce is rent, not ownership.
Move the Boundary to the Key
If the law will not enforce ownership before the copy, enforcement has to live where it cannot be argued away: in the key that unlocks the work. That is what decentralized DRM does, and it is the opposite of the DRM you learned to resent, because it serves the owner instead of the store. The party who receives the work gets the experience it needs, not the source it could keep.
With Elacity dDRM, your work stays encrypted everywhere except one sealed moment of use inside a locked sandbox. The key that opens it is split across independent machines, assembled to decrypt for a single authorized action, then wiped. No app, no platform, and no model ever holds it. Elacity's own strongest claim is that keys are used, never owned.
Package the work as a Wealth Capsule and the terms travel with it. Every attempt to open it re-checks your rights on-chain and honors the royalty you wrote in. An AI that wants your work meets the same gate as a person: prove the right, pay the terms you set, receive the output it needs, and never touch the raw file.
- The work is sealed before it leaves you, not tracked after it spreads.
- Access is a permission you grant and can revoke, not a copy you hope no one abuses.
- The terms are enforced by the key at the moment of use, not by a court years later.
- You set the price and the royalty, and the work carries both wherever it travels.
What This Does Not Fix
Precision matters. This does not repeal fair use, and it cannot recall a file already sitting inside a training set. The fully autonomous economy, where agents transact and pay per use as they work, is a direction Elacity is building toward, not a finished storefront. What ships today is the harder primitive underneath it: work that can be used without being surrendered.
The point is not to punish AI for wanting your work. It is to make wanting it mean paying for it on your terms. Elacity's founder, Sasha Mitchell, states the principle without hedging:
The people who create the value should own it. That is the entire reason Elacity exists.
The government just told creators to change the law if they expect to be paid. That fight is worth having, and it will take years. While it runs, you can change the file: seal the work, set the terms, and make every copy meet a gate it cannot talk its way past.
For more on how policy keeps landing a step behind ownership, read the Ecosystem and Governance hub. When you are ready to stop leaning on the courtroom, follow Elacity on X.