← All articles

Mikhail Polyakov

USA Today seeks more than $250 million from OpenAI: what the dispute is about and what a company using neural networks should check

Level: BeginnerReading time: 7 min

Topics: AI, Small business

A stack of newspapers with no readable text on the left, a glowing network of dots and lines on the right, balanced scales of justice between them

Short answer: on 8 October 2026 USA Today Co. and several local newspapers it owns filed a lawsuit against OpenAI. The plaintiff claims that "hundreds of thousands" of its articles were copied to train models without permission, and seeks damages of more than $250 million. These are the plaintiff's claims, not established by a court: the case has only been filed, and there is no ruling. According to The Verge, there are already many copyright cases against OpenAI, from The New York Times lawsuit to a coalition of nearly 400 local newspapers. What exactly is sought in each of them cannot be seen from that source. For a company that simply uses a chatbot, the practical meaning of this news comes down to two questions of its own: where the material you publish came from, and what your subscription terms say about responsibility. Below: what is in the lawsuit, who else is in court, the difference between training and output, and what to check at your end.

What exactly happened

According to The Verge's report of 8 October 2026, USA Today Co., together with several of its local newspapers, is suing OpenAI. The claim: the company copied "hundreds of thousands" of articles in order to train its AI models on them. The case, The Verge writes, was reported earlier by Reuters.

In a filing on Thursday, the publisher asks for damages of more than $250 million and alleges that the unauthorized use of its content "has done real and continuing" harm to its outlets. Other wordings from the same place: "OpenAI's commercial success rests on large-scale copyright infringement"; permission was never asked of the publisher, and the lawsuit alleges that OpenAI instead "took" its content to "build products worth hundreds of billions of dollars".

All these words matter as the plaintiff's position. They have not been established by a court. OpenAI, according to The Verge, did not immediately respond to a request for comment.

Besides USA Today itself, the lawsuit includes The Tennessean, Indy Star, The Columbus Dispatch, The Oklahoman and several other local outlets.

I was not able to obtain the text of the lawsuit itself, so everything said above about its content comes from The Verge's account, which cites Reuters.

A string of copyright cases

The Verge describes the lawsuit as the latest in a string of copyright cases against OpenAI. Besides The New York Times lawsuit, the outlet names legal action from:

  • The Intercept;
  • Ziff Davis, the owner of CNET;
  • CBC/Radio-Canada;
  • Encyclopaedia Britannica;
  • Merriam-Webster;
  • The Seattle Times;
  • a coalition of nearly 400 local newspapers.

This is a list as of 8 October 2026 according to one outlet, not a complete register of all cases. How each of them will end cannot be said from this material.

What can be seen from this list without guesswork: The Verge names these organisations among those bringing cases against OpenAI. The subject matter of each of these cases cannot be established from this source, while the USA Today case, by the outlet's account, concerns the use of articles for training models.

How a dispute about training differs from a dispute about output

In this case, by The Verge's account, the claim is specifically about copying articles for training. That is not the same as a claim about a particular chatbot answer that retold someone else's story. For a user the difference is practical.

Training. This is the stage at which a model is prepared, and it happens at the tool provider's end, before you ever opened a chat window. The composition of the training material was not chosen by you, and in the case described it is precisely that which became the subject of the dispute. In the case described, the dispute about this stage is between a publisher and a model developer; users of the tool are not party to it.

Output. This is the text, image or code you received in an answer and then publish on your website, in a newsletter or in advertising. The decision to publish is yours, regardless of who or what prepared the text, and it is an editorial decision of your own.

The lawsuit described concerns the use of articles for training models and says nothing about the legal position of other ways of working with such tools. So what is worth going through is your own part: what exactly you publish and where it came from.

What to check at your end

Four checks. None of them requires a lawyer on an ongoing basis; each has someone responsible and a sign by which you can see it has been done.

  1. Read the terms of your own tool and your own plan on rights to the output and on responsibility. What they say cannot be learned from the news: it has to be looked at in the documents of the service you use, and on the plan you pay for.
    • Who does it: whoever pays for the subscription, together with a lawyer for a single review.
    • How to check: you have an extract from the terms as of the date of the check, showing who owns the output, what prohibitions there are on its use, and whether the provider takes on anything in case of third-party claims. If you could not find the relevant terms, write that down as an open question and put it to the service's support in writing. Until there is an answer, such material is not published anywhere: not on the site, not in a newsletter, not in advertising. Instead you take material whose terms of use are clear.
  2. Know the provenance of the material you publish. The practical difference is whether you can name the source of every figure and every quotation in your text. The way the text was written does not settle that in itself: it has to be checked against the text, and material for which sources cannot be named goes back for rework.
    • Who does it: whoever runs the site or the newsletter.
    • How to check: for every publication over the last month you can name the source of its figures and quotations. If there is no source, the material goes back for rework.
  3. Set a rule for quotations and images. For every third-party photograph and every long quotation you need to know the source and the terms on which it may be used.
    • Who does it: the editor or whoever publishes.
    • How to check: the drafts show that every quotation has its source given, and that for every image it is known where it came from and on what terms it is used. If the rights to an image remain unclear, the material is not published with it: you use another image with clear terms, or publish without one.
  4. Keep drafts and your exchanges with the tool for important material. This is so that for any publication you can reconstruct where its figures, quotations and images came from.
    • Who does it: whoever prepares the material.
    • How to check: for any publication over the last six months you find the draft and the sources within a minute.

How I set up checking material before publication, with mechanical checks and independent proofreading by another model, is described in the article Automating content creation with AI. How the story of streaming fraud with AI songs ended is covered here: Streaming fraud with bots and AI songs.

What not to expect from this news

  • A conclusion about who is right. The lawsuit has been filed, the defendant's position is not given in the report, and there is no ruling.
  • Transferring the sum to your own case. More than $250 million is the plaintiff's claim in this particular case. How the sum was calculated is not explained in the available account, and it does not set a general monetary benchmark for disputes of this kind.
  • Rules for Russia. Russian rules on rights to works are not examined in this article, and the conclusions here do not rest on them. Which court is hearing the case and under which law is not stated by the available source.
  • A ban on the tools. The available account reports a monetary claim. What other claims the lawsuit itself contains cannot be established from this source.

In summary

On 8 October 2026 USA Today Co. and several of its local newspapers filed a lawsuit against OpenAI, seeking more than $250 million and claiming that hundreds of thousands of their articles were copied to train models. This is the plaintiff's claim, and the case has only begun. According to The Verge, there are already many copyright cases against OpenAI, but their subject matter cannot be established from that source. For a company that uses a chatbot in its work, it is more useful not to follow someone else's proceedings but to take care of what is within your control: read the subscription terms on rights to the output, know the provenance of what you publish, put quotations and images in order, and keep your drafts.

I work on AI agents and automation. If you would like to see my projects or discuss your own task, take a look at my portfolio.

Sources