polski
español
A small publisher can be two things at once.
It can be a customer using artificial intelligence to reduce repetitive production work.
It can also be a supplier whose reporting adds valuable knowledge to an AI product.
Buying the first service should not require the publisher to surrender its bargaining position in the second market.
That distinction is missing from an unusual argument Washington has presented in one of the most consequential copyright cases of the AI era.
What the Justice Department told the court
On September 1, 2026, the U.S. Department of Justice filed a statement of interest in the consolidated copyright litigation involving OpenAI, Microsoft, The New York Times, book authors and other publishers.
The plaintiffs allege that copyrighted books and news articles were used without permission to train OpenAI’s models. OpenAI and Microsoft argue that the training is protected by fair use.
Under Section 107 of U.S. copyright law, fair use is assessed through four factors, including the purpose of the use, the nature and amount of the copyrighted material and the effect on the potential market for the original work.
The Justice Department’s filing supports the technology companies’ interpretation. It argues that training language models is highly transformative because the resulting systems perform functions different from those of the individual works used during training.
The government also frames American AI development as an economic and national-security priority. It warns that an unfavorable ruling could raise development costs, weaken smaller AI companies and give foreign competitors an advantage.
This is the administration’s position, not a court ruling.
Both sides have asked U.S. District Judge Sidney Stein to rule in their favor on key issues without a trial. As of this writing, the court has not decided whether OpenAI’s training practices qualify as fair use.
Washington’s argument about small publishers
The filing makes a second argument aimed specifically at the publishing market.
If AI companies must purchase licenses for training material, the government says, publishers with the largest archives could receive the greatest payments. Licensing costs could become a barrier that only the largest technology companies can afford, while large legacy media organizations collect most of the money.
At the same time, the filing argues that AI helps underfunded newsrooms compete.
One example is a local outlet using an image generator to create an illustration that might otherwise require a photographer or a licensed image. Other AI tools can assist with translation, transcription, research organization and routine editing tasks.
The concern about market concentration is legitimate.
Current AI licensing deals have largely favored national and international publishers with large archives, recognizable brands and the legal resources to negotiate. A licensing system designed only around volume could leave a weekly newspaper outside the room.
But that is an argument for building a licensing market that includes small publishers.
It is not an argument that their reporting has no compensable value.
The illustration is not the expensive part
An AI-generated illustration may lower the cost of presenting a story.
It does not produce the underlying journalism.
Someone still has to attend the school board meeting, read the budget, request the contract, call the superintendent and give every party an opportunity to respond. The newsroom pays for that work before there is anything for an AI model to summarize, analyze or learn from.
For a local newspaper, the expensive input is often not the image beside the article. It is the reporter’s time and the institutional knowledge required to understand what happened.
That reporting may also be unusually valuable to an AI system precisely because it cannot be found elsewhere.
A large national publisher offers scale. A local newspaper offers facts that may exist nowhere else: the result of a county vote, the name of a new school superintendent, the terms of a zoning decision or the history behind a local dispute.
Archive size is therefore an incomplete measure of licensing value.
Ten thousand versions of information available throughout the internet may contribute less unique knowledge than one original report from a town nobody else consistently covers.
Being an AI customer does not eliminate supplier rights
The government’s argument treats AI assistance and compensation for training material as if they were opposing interests.
They are not.
A farmer can buy a tractor and still charge for the crop. A photographer can use editing software and still license the finished image. A publisher can use an AI tool and still negotiate the terms under which its journalism contributes to someone else’s commercial product.
CMS4media’s AI Assistant, for example, can help editors prepare titles, introductions and SEO descriptions from reporting already entered into the system. It can reduce repetitive work while leaving factual verification and final editorial judgment with the newsroom.
That is a customer relationship. The publisher uses a tool to improve its own workflow.
Training or grounding an external AI product with that publisher’s reporting creates a different relationship. The AI company is using an input created and financed by the newsroom.
The fact that the publisher benefits from AI in one context does not settle what should happen in the other.
Affordable tools and fair compensation can coexist comfortably.
A court victory would not automatically help every weekly newspaper
The opposite oversimplification should also be avoided.
If the court rejects OpenAI’s fair use defense, that would not automatically generate a check for every local publisher whose work may have entered a training dataset.
Small outlets would still face practical barriers:
- determining whether their content was used;
- finding the party responsible for that use;
- negotiating without dedicated legal staff;
- understanding the rights covered by an agreement;
- verifying usage and payments;
- collecting sums too small to justify individual enforcement.
The experience of other media-support systems shows why distribution matters as much as the headline amount. As we observed in Australia Can Make Big Tech Pay. It Cannot Make Big Tech Bargain, a policy can require technology platforms to contribute without guaranteeing that independent local publishers receive meaningful individual agreements.
The same risk exists in AI licensing.
If each newspaper must negotiate separately with a global technology company, the largest publishers will continue to have the advantage. A workable system for local news may require collective negotiation, standardized contracts, transparent usage reporting or a clearinghouse that can distribute smaller payments efficiently.
The important question is not only whether licensing is required.
It is whether a local newsroom can participate without spending more on lawyers and administration than it receives for its work.
What fair participation could look like
A system designed for small publishers would need to reduce both the power imbalance and the transaction cost.
That could include:
- clear information about whether a publisher’s content was included in training or used to ground an answer;
- standard definitions of training, retrieval, grounding, citation and display;
- contracts written for publishers without in-house legal departments;
- collective or representative negotiation;
- payments that recognize original and locally exclusive reporting, not only archive volume;
- an accessible method for opting in, opting out or reserving specific rights;
- reporting that allows publishers to verify how compensation was calculated;
- a practical dispute process for missing or incorrect payments.
None of those mechanisms requires the court to decide that every use of copyrighted material is infringement.
Fair use is context-specific. Different methods of acquiring, storing, transforming and reproducing material may produce different legal outcomes.
But the government’s small-publisher argument does not resolve those questions. It simply assumes that access to inexpensive AI tools may be more valuable to a small newsroom than the ability to negotiate over its own reporting.
Publishers should not have to accept that trade without evidence.
Small publishers need a seat, not an exemption from payment
The Justice Department is right about one risk: a licensing market can be designed badly.
It could reward archive size, exclude new entrants and concentrate both payments and AI development among a handful of large companies.
The solution is not to pretend that small publishers benefit when their bargaining rights disappear.
A local newsroom can use AI to work more efficiently and still expect fair terms when its original reporting helps build another company’s product. It can welcome cheaper production tools without treating the reporter’s labor as a free input.
The real conversation should focus on participation.
How can a weekly newspaper know when its work has been used? How can it negotiate without mounting an expensive lawsuit? How should payments recognize unique local reporting instead of rewarding volume alone?
A cheaper illustration may help publish the story.
Someone still has to pay for the reporter who discovered there was a story to tell.
