Why USA Today Is Suing ChatGPT's Maker for $250 Million

USA Today Co. is suing OpenAI, saying the company copied hundreds of thousands of its news articles to teach its AI systems without permission. The publisher wants more than $250 million in damages The Verge.
The amount is clear. The claim is simple. According to the complaint, OpenAI never asked to use the content.
The case covers more than USA Today. It includes The Tennessean, Indy Star, The Columbus Dispatch and The Oklahoman. The claim is not about one copy or one answer. It says OpenAI regularly took large amounts of fresh news to teach general AI systems, a bit like letting a student read a huge library to learn how to write.
USA Today Co. is not alone. The New York Times sued OpenAI and Microsoft in 2023, saying they used millions of its articles. Microsoft is OpenAI's largest financial backer Reuters.
The Seattle Times and Newsday also sued OpenAI and Microsoft for alleged copyright infringement. That case was filed in the U.S. District Court for the Southern District of New York. It says the companies copied material from the newspapers' websites Reuters.
More cases keep coming. More than 30 U.S. local newspaper groups that together own about 400 titles have sued OpenAI and Microsoft over unauthorized copying and use Press Gazette. An earlier case from eight newspapers accused OpenAI, the company behind ChatGPT, of taking millions of news articles without permission or payment.
The dispute is not only about newspapers. Encyclopedia Britannica said OpenAI unlawfully copied nearly 100,000 of its articles to train GPT models. Major publishers have separately said Meta used millions of books and articles without permission to train its Llama model. Authors suing Microsoft over a model allegedly taught on pirated books asked for a court order to stop the copying and up to $150,000 for each work allegedly misused.
Across these cases, the tech questions are the same. How was the text collected. Whether collecting from the web, preparing the data, breaking text into small pieces and adjusting internal settings counts as copying under copyright law. Where the line is between learning from data and repeating it word for word when answering. The answers depend not only on how the AI is built but on records showing where data came from.
The broader context here is that courts are being asked to set rules for the supply of data behind large AI systems. Publishers want prior permission and payment. Developers point to fair use, a legal rule that allows limited use of copyrighted work, and say training transforms the material.
In my view, that either-or argument hides the more lasting business and engineering problem.
For context on how we got here, the web was built for search and reading, with robots.txt, a simple file that tells automated programs what they can copy, and terms of service as imperfect stands-ins for permission. Large-scale training broke those assumptions. It needs clear permission signals that machines can read and licensing that works at web scale.
Looking at what this means for buyers and builders, the near-term risk is less about one verdict than split rules. Different publishers, places and payment demands will apply. Business teams already track data rules, model documentation and legal protections. That work will now have to cover where training data comes from and how outputs are filtered.
The long-term picture here remains hopeful. Licensed access to good reporting, reference and book content would make AI better. It would cut down on false answers, improve sourcing and pay for original work. My children grew up moving from printed encyclopedias to Wikipedia to chatbots, and each step made facts easier to find and harder to trace. Fixing that trace is where the real opportunity lies.


