USA Today sued OpenAI for more than $250 million. That number is a demand, not a price.

What happened

USA Today Co. sued OpenAI in Manhattan federal court on Thursday. The complaint alleges OpenAI copied hundreds of thousands of articles to train its commercial models without permission. The publisher is seeking more than $250 million in damages plus a court order.

That's the confirmed part: named plaintiff, named defendant, court, filing date, alleged scale, dollar figure.

It lands the same stretch in which more than 300 publishers — Condé Nast and Hearst among them — are pressing Congress for federal action against AI scraping, according to reporting on that lobbying push. Courts on one track, legislation on the other.

Why the number matters, and how it doesn't

$250 million is a damages demand written by the plaintiff's lawyers, not a valuation of a news archive by anyone with the power to set one.

That distinction is the whole story. A complaint can ask for any figure. A court decides what, if anything, gets paid. Settlements routinely land somewhere unrecognizable from the opening number, and many of these cases never reach a published figure at all.

What it does do is move the anchor. Until now, the public numbers in this market have been small and oddly specific: $5,000 for a book in a training deal, $1 per 1,000 tokens for retrieval of peer-reviewed research, undisclosed per-value payments from Google. A nine-figure demand from a top-tier US news publisher sits in a different register. Anyone across a table from an AI company's business development team now has a louder number to point at — even if it's unproven.

What this is NOT

It is not a ruling. Nothing here establishes that training on published articles is infringement. That question is still open, and an appeals court has only partly addressed it.

It is not an injunction. The available reporting does not confirm USA Today sought or obtained any order affecting OpenAI's products right now. Nothing you use today stops working.

It is not a payout. No settlement has been announced. No money has moved.

It is not a precedent for you. USA Today owns a large, registered, clearly documented archive with institutional copyright hygiene. A solo newsletter writer or a YouTuber with video transcripts is in a different evidentiary position — not a worse one, but a different one.

And the exact article count, the specific datasets, the model versions, and whether OpenAI outputs reproduced USA Today text are all unconfirmed in the available reporting.

The question underneath

Can a publisher stop an AI company from training on its work?

Filing suit challenges the use. It does not automatically halt training and it does not guarantee payment. The outcome turns on who owns the copyright, what the court decides about training as a use, what the evidence shows about copying and outputs, and whether any contract or terms already covered access.

That's why the timing question matters more than the headline. A publisher who signs a broad agreement this quarter may be trading away leverage that gets more valuable if OpenAI's defenses fail — or locking in revenue that looks smart if they hold.

What to do about it

1. Pull your server logs now. Preserve evidence of AI crawler traffic and any instance of your text appearing verbatim in a chatbot answer. Screenshots with dates.
2. Confirm your copyright registrations. Statutory damages in the US generally require timely registration. If your archive isn't registered, that's a conversation with counsel this month, not next year.
3. Separate archive rights from live-web access in any licensing talk. They are two products. Price them separately.
4. Do not quote $250 million, or any fraction of it, as a benchmark. It is an unadjudicated demand from a publisher with an enormous back catalog.
5. Re-read your robots.txt and crawler terms. Know what you have already permitted before you claim it was taken.
6. If you're mid-negotiation, ask in writing whether the deal releases past claims. That clause is where the leverage goes.
7. Watch for OpenAI's answer to the complaint. That filing will show which defenses it leads with — and that tells you more about your own position than the damages number does.