15 Oct
2026
In roughly five weeks, the fight over how artificial intelligence companies use journalism stopped being a single fight. It is now running on three fronts at once: in courtrooms, in licensing offices, and inside a fast-growing alliance of publishers trying to write the rules before someone else writes them. The latest filings and deals, logged in Press Gazette's running tracker of publisher AI lawsuits and agreements, show an industry that has not chosen between suing and signing. Increasingly, it does both.
The underlying question has shifted. Few people still argue about whether large AI systems learned from news content. The live questions are who gets paid for that, how much, and, just as important, who gets to measure the use in the first place.
On October 9, USA Today Co sued OpenAI over the alleged unauthorized use of "hundreds of thousands" of articles from its 19 publications, seeking $250 million in damages, according to Press Gazette. The suit names titles well beyond the flagship, including The Tennessean, IndyStar, the Detroit Free Press and The Des Moines Register.
What stands out is the counting. The complaint says OpenAI's WebText dataset, used to train GPT-2, included more than 160,000 entries from USA Today Co's newsbrands, 83,266 of them from USA Today itself. It says a 2019 Common Crawl snapshot used to train GPT-3 contained more than 122 million tokens from its publications. And it alleges the models have "memorised" its work and can produce "near-verbatim copies and summaries" when prompted. These are allegations, not findings, and OpenAI will have its chance to contest them. But a complaint built on inventories rather than general grievance tells you where the litigation is heading: toward specific, auditable claims about what went in and what comes out.
USA Today Co is not alone. On September 7, The Seattle Times and Newsday sued OpenAI and Microsoft over alleged scraping and use of their content, including paywalled articles. "We feel strongly that we must defend our content," Seattle Times CEO Alan Fisco said. Regional and local publishers, which often lack the leverage to negotiate licensing deals on favorable terms, appear increasingly willing to let a court set the price instead.
The legal backdrop moved in late September. The U.S. Court of Appeals for the Third Circuit upheld Thomson Reuters' win against Ross Intelligence, a now-defunct legal AI startup, as Nieman Lab reported. It is the first federal appeals ruling to reject a fair use defense for using publisher content, without permission or payment, to train a commercial AI product. Judge Tamika Montgomery-Reeves wrote that the dispute was "no more than an ordinary copyright case."
The case centered on thousands of Westlaw "headnotes," the short case summaries Thomson Reuters' editors write, which Ross allegedly copied to build a competing legal search tool. The court found the material had a "creative spark" and that Ross used it for a highly similar purpose in direct competition.
Publishers should not over-read the result. Ross's tool retrieved existing passages from judicial opinions; it did not generate new text, and the case does not involve a large language model. The court explicitly distinguished Bartz v. Anthropic, where a judge found that training on published books was fair use and where liability instead turned on pirated copies, a case that ended in a $1.5 billion settlement. The opinion also leaves room for generative tools to be found more transformative than a search engine. In other words, the ruling strengthens the hand of rights holders whose work is used to build a market substitute, but it does not decide the generative AI cases now piling up.
While some publishers sue, others sign. On October 12, Reach joined Google's AI news pilot, giving Google "enhanced content rights" across its UK and Irish portfolio, including the Mirror, Express, Daily Star, Manchester Evening News and Liverpool Echo. Google describes the arrangement as a commercial partnership in which it pays for those rights and tests experimental features; it has said it is trialing AI-generated article overviews on participating publishers' Google News pages, along with audio briefings. No payment figures were disclosed. Reach CEO Piers North said the company's content offers something readers "still value in the age of AI."
OpenAI has also been widening its partner list. In September it signed its first Canadian partnership, with Village Media, providing funding, API credits and technical support for a local health navigator app in exchange for Village Media content being cited in ChatGPT. Later that month it signed deals with The Times of India's publisher, Bennett, Coleman and Co, and the Indian Express Group; trade outlet e4m reported they could be worth around $5 million and $3 million a year respectively, figures that remain unconfirmed.
It is tempting to frame this as a split between hawks and doves. A better reading is that lawsuits and licenses are two ways of doing the same thing: putting a price on news as an input. A lawsuit seeks a price set by a court. A license seeks a price set by negotiation. Each makes the other more credible.
The most consequential development may be the least dramatic. On October 13, Press Gazette reported that more than 40 new members had joined SPUR, the Standards for Publisher Usage Rights coalition, including The New York Times, The Atlantic, The Economist, ITV News and The Observer. SPUR now counts 80 publishers and publishing associations from more than 20 countries. It was launched in February by The Guardian, the Financial Times, The Telegraph, the BBC and Sky News, with Mediahuis, the Associated Press and CMA Media later added as founding members.
Earlier in October the coalition released an open-source content telemetry standard designed to track how AI agents use publisher material, covering scraping, grounding, citation, on-screen display and user engagement. "Intellectual property protection and technological advancement can and must co-exist," said Rebecca Grossman-Cohen, a senior vice president at The New York Times. SPUR co-founder David Buttle put it more bluntly: "We can't simply hope for the best when it comes to AI companies and the future of publishing."
Every licensing negotiation runs into the same wall: the publisher usually cannot see how its content is being used once it enters an AI system. Without that visibility, deals default to flat fees that reflect bargaining power rather than value. A shared telemetry standard, if AI companies adopt it, would let publishers argue from usage data instead of guesswork, much as music rights organizations argue from play counts.
That is a big "if." Adoption is voluntary, and the companies with the most data have the least incentive to share it. But the combination of rising litigation, an appeals ruling that rejects fair use in at least one commercial context, and an 80-member coalition asking for the same thing gives publishers more collective leverage than they have had at any point since generative AI went mainstream.
The news business has spent two decades negotiating with platforms from a position of weakness. This round is different in one respect: publishers are, for once, coordinating early. Whether that turns into fair compensation will depend less on any single verdict than on whether the industry can make AI usage as visible as a page view.
A message can lift you, but the right fit helps you wear it your way. Choose an uplifting women's...
A September research announcement revisits BASE-STEP’s March 2026 antiproton transport. The scien...
Ahead of November 3, UNESCO’s 2026 designations offer a way to read conservation, livelihoods and...
Account
Cart