PolicyLaw, regulation & governance
USA Today Co. sues OpenAI for more than $250 million and asks the court to order GPT models destroyed
USA Today Co. and its local papers say OpenAI copied hundreds of thousands of their articles to train ChatGPT. In federal court they seek more than $250 million and the destruction of the GPT models and training sets built on that work.
Reach: not measured yet
AT Pulse research deskPublished · 5 min read · AI-assisted analysisThe publisher's biggest demand is not the money. It wants the court to order the destruction of the GPT models and training datasets built from its journalism. Seattle Times and Newsday asked for the same thing in their suit a month earlier. The more-than-$250 million figure is a demand in a court filing, not an award. [1] [3] [7]
The complaint makes specific claims about training data. It says USA Today content accounts for more than 160,000 WebText entries and more than 122 million tokens in a 2019 C4 snapshot. It also argues that ChatGPT's outputs can replace a visit to the original article. [3] [8]
The case joins a large group of consolidated publisher suits against OpenAI before Judge Sidney Stein in Manhattan. No court has yet ruled on fair use for news training. In September, the Justice Department filed a non-binding statement arguing that LLM training can be fair use. [1] [5] [9]
Statutory damages are counted per work and can reach $150,000 per work for willful infringement, so large news corpora carry large legal exposure. That makes licensing more attractive. OpenAI already has publisher deals, such as with News Corp, and the DOJ suggested Congress consider licensing systems. [3] [5] [8]
What is known: the suit has been filed and its demands are public, but no court has decided fair use for news training.
Read the full assessment
What may follow: if demands to destroy models gain traction, tracking data sources and licensing content become day-to-day requirements, not just legal precautions.
Executive brief
USA Today Co. is asking a court to order the destruction of GPT models and training sets built on its journalism. That remedy matters more than the headline figure of more than $250 million in damages (Hoodline). The publisher and its local papers sued OpenAI in Manhattan federal court on October 8, 2026. They say OpenAI copied "hundreds of thousands" of articles to train ChatGPT (The Verge). The case joins a large group of publisher suits. It comes a month after the U.S. Justice Department told the same court that AI training can be fair use. The $250 million is what the publisher is asking for, not an award.
What changed and event timeline
Anthropic's book settlement approved
A judge gave final approval to Anthropic's $1.5 billion class settlement over pirated books, about $3,100 per work. It is the largest copyright class settlement in U.S. history ().
DOJ backs OpenAI
The Justice Department filed a non-binding statement of interest in the consolidated OpenAI litigation, arguing that LLM training can be fair use (). TheWrap dated the filing a day later ().
Seattle Times and Newsday sue
They sued OpenAI and Microsoft in the Southern District of New York (S.D.N.Y.), alleging paywalled content was scraped. They also want the models destroyed ().
USA Today Co. files
The complaint seeks more than $250 million and names The Tennessean, IndyStar, The Columbus Dispatch, The Oklahoman and other papers (). Hoodline lists the filing date as October 9 ().
Capabilities and access
This is a lawsuit, not a product launch. The complaint targets ChatGPT and the GPT models behind it. The reviewed coverage does not name specific model versions (RuntimeWire). Coverage says it covers work from 19 publications, including USA Today, the Detroit Free Press, The Arizona Republic and the Milwaukee Journal Sentinel. No injunction has been issued, and ChatGPT access is unchanged.
Technical analysis for researchers and developers
- Dataset-level evidence: Hoodline reports the complaint's allegations that USA Today content makes up more than 160,000 entries in WebText, plus more than 122 million tokens in a 2019 Common Crawl C4 snapshot, 23 million of them from usatoday.com (Hoodline).
- Output claims: The complaint says ChatGPT can reproduce or summarize articles closely enough to replace a visit to the source (RuntimeWire).
- Removed attribution: It also seeks up to $25,000 per instance of removed copyright management information, meaning stripped bylines and copyright notices.
Read the full section
- Dataset-level evidence: Hoodline reports the complaint's allegations that USA Today content makes up more than 160,000 entries in WebText, plus more than 122 million tokens in a 2019 Common Crawl C4 snapshot, 23 million of them from usatoday.com (Hoodline). These are allegations about dataset composition. Such counts do not show what is in current production training mixes.
- Output claims: The complaint says ChatGPT can reproduce or summarize articles closely enough to replace a visit to the source (RuntimeWire).
- Removed attribution: It also seeks up to $25,000 per instance of removed copyright management information, meaning stripped bylines and copyright notices. That puts scraping and cleaning pipelines under scrutiny.
Claims and evidence
- More than $250M and model destruction sought: reported by several outlets (Hoodline; The Verge). These are demands in a filing, not findings.
- Token counts: these come only from the complaint, as relayed by Hoodline. No independent audit of the datasets was found.
- Plaintiff count conflicts: Hoodline says USA Today Co. "and 13 sister newspapers," while RuntimeWire describes 19 owned publications.
Read the full section
- More than $250M and model destruction sought: reported by several outlets (Hoodline; The Verge). These are demands in a filing, not findings.
- Token counts: these come only from the complaint, as relayed by Hoodline. No independent audit of the datasets was found.
- Plaintiff count conflicts: Hoodline says USA Today Co. "and 13 sister newspapers," while RuntimeWire describes 19 owned publications. The 19 likely counts publications whose work appears, not named plaintiffs.
- OpenAI's position: OpenAI says training on publicly available material is fair use (RuntimeWire). It had not commented on this specific suit.
Context and prior work
Read the full section
- The New York Times sued OpenAI and Microsoft in 2023. Publisher cases are now consolidated as In re OpenAI Copyright Infringement Litigation, No. 25-md-03143, before Judge Sidney Stein (Bloomberg Law).
- In March 2025 the judge let the Times case proceed, with fair use still undecided (NHPR/NPR).
- Other plaintiffs include The Intercept, Ziff Davis, CBC/Radio-Canada, Britannica and a coalition of nearly 400 local papers (The Verge).
- OpenAI has signed licensing deals with some publishers, such as News Corp in 2024 (RuntimeWire).
Limitations, safety and contested findings
- No court has ruled on fair use for LLM training in these news cases.
- The DOJ statement carries no binding authority (Bloomberg Law). The Times called it siding with "trillion-dollar AI companies" (TheWrap).
- In July 2026, the Times sought sanctions, saying OpenAI withheld evidence in discovery (TheWrap).
Read the full section
- No court has ruled on fair use for LLM training in these news cases.
- The DOJ statement carries no binding authority (Bloomberg Law). The Times called it siding with "trillion-dollar AI companies" (TheWrap).
- In July 2026, the Times sought sanctions, saying OpenAI withheld evidence in discovery (TheWrap).
- Anthropic's settlement turned on pirated books, so it does not settle questions about scraping the open web.
Business and practitioner implications
- Model-destruction demands are now standard in publisher suits (Seattle Times, Newsday, USA Today).
- Willful infringement can bring up to $150,000 per work (Hoodline), so corpora with many articles carry large exposure.
- Licensing pressure: the DOJ itself suggested Congress look at licensing or collective-rights systems (Bloomberg Law). Licensing content is the lower-risk path.
Read the full section
- Model-destruction demands are now standard in publisher suits (Seattle Times, Newsday, USA Today). Teams should keep records of where their data came from so specific sources can be traced and removed.
- Statutory damages scale per work. Willful infringement can bring up to $150,000 per work (Hoodline), so corpora with many articles carry large exposure.
- Licensing pressure: the DOJ itself suggested Congress look at licensing or collective-rights systems (Bloomberg Law). Licensing content is the lower-risk path.
- Output controls: filters that limit verbatim reproduction and summaries that replace the source address the market-substitution argument directly.
The source trail.
Sources (9)
USA Today becomes the latest publisher to sue OpenAI
Article text retrieved; extracted text may omit tables or interactive elements.
theverge.comUSA Today Co. sues OpenAI for more than $250M over use of news articles in training
Related coverage; assess separately
contentgrip.com