Sofia follows the money, policy, and platforms shaping what creators can make.
Get the week's top AI and AI-art stories delivered to your inbox — curated, concise, free.
Free. Unsubscribe any time.
Pick a companion and get their take on this story
USA Today Co. and several of its local newspaper properties are suing OpenAI for more than $250 million, alleging the company scraped and copied hundreds of thousands of their articles without authorization to train its AI models — the latest in a string of major publisher lawsuits that are steadily narrowing the legal ground under every AI text model on the market.
The damages figure is not incidental. When the New York Times sued OpenAI in late 2023, it did not specify a dollar amount — it sought "billions." USA Today Co.'s explicit $250 million ask is a tactical choice: a concrete, defensible number tied to the scale of alleged copying. It gives a jury something to anchor on and signals the publisher has done the math on licensing value foregone.
The suit covers not just the flagship USA Today masthead but the local newspapers USA Today Co. owns — a detail that matters because local reporting has historically been the unglamorous backbone of training corpora. It is precisely the kind of regional, high-volume, factual text that makes AI outputs sound grounded and authoritative. Publishers know this, and they are increasingly pricing it accordingly.
The New York Times, the Chicago Tribune group, and now USA Today Co. — the list of litigants is long enough that "isolated dispute" is no longer a credible framing. Each new filing adds to the evidentiary record that the industry-wide practice of scraping publisher content was not a legal grey area that reasonable actors could navigate in good faith; it was a calculated bet that the cost of litigation would be lower than the cost of licensing.
That bet is looking shakier by the quarter. OpenAI has signed licensing deals with some publishers — the Associated Press and Axel Springer among them — but the terms have not been disclosed, and the existence of paid deals arguably undermines the fair-use argument for unpaid scraping. If licensing was available, why wasn't it used?
For context, Stability AI faced similar pressure in 2023 when artists and Getty Images filed suit over image training data — a dispute that forced the company into years of legal uncertainty and contributed to its leadership churn. The trajectory for text models now rhymes closely with that earlier image-data fight.
The practical consequence for AI-art creators is indirect but real. The text models that power prompt interpretation, character description, and style guidance in tools like ChatGPT — which feeds into image-generation workflows across the industry — are built on the same training pipelines now under legal scrutiny. If courts begin issuing injunctions or forcing data removal, the next generation of multimodal models could arrive with deliberately impoverished text corpora, affecting how well they parse nuanced creative prompts.
OpenAI's own recent moves around EU watermarking and compliance suggest the company is already operating in a more legally cautious posture. Copyright litigation is another dimension of that same pressure. The question is whether legal settlements will produce licensing frameworks that stabilize the training-data supply chain — or whether they will fragment it in ways that advantage well-capitalized players who can afford blanket deals and leave smaller model developers unable to compete.
Sen. Adam Schiff's recent assessment of AI copyright gridlock in Congress makes clear that legislative relief for AI companies is not coming soon. That leaves the courts as the primary arena, and USA Today Co.'s $250 million filing is the latest evidence that publishers intend to use them aggressively.
The next concrete milestone to watch: whether OpenAI moves to dismiss or attempts to settle before discovery begins — because discovery, which would force disclosure of exactly what data was used and when, is what the company has most to lose from.
