Paywall violation ‘eviscerates’ fair use defense: What OpenAI-NYT twist means for publishers

The filings, unsealed last Thursday, show executives talking about the “gazillions” of dollars at stake in commercializing AI models while privately acknowledging the “existential threat” those products pose to publishers’ underlying economics. 

It’s the clearest public window yet into what tech executives were saying internally about the issues at the heart of the case. And if the court buys this picture, it doesn’t just ding OpenAI and Microsoft; it reshapes the terms of trade between AI and publishing, and could hand publishers their strongest leverage yet over training data and licensing. 

Digiday spoke to a range of lawyers and AI licensing experts about what the filings signal for publishers’ AI monetization strategies. 

‘Eviscerates’ the fair-use argument for training on news 

Legal and regulatory experts say the newly unsealed documents are devastating for OpenAI and Microsoft’s fair-use claim over training on news.

The fourth fair-use factor is the impact on the market for the work. The filings show a real licensing market already exists and was operating inside the defendant’s partnership. Content was explicitly priced and traded between Microsoft and OpenAI, and all the major AI players now sign publisher licensing deals for similar uses, stressed David Buttle, founder of publisher AI licensing coalition Spur.

“OpenAI, Microsoft, Amazon, Google, Meta, Perplexity and Mistral have all signed publisher licensing deals covering the same content and uses,” he said. “The defendants’ own conduct establishes that the content has a market price. The case is about their refusal to pay it to these plaintiffs.”

Microsoft CEO Satya Nadella’s acknowledgment that AI chatbots can provide information from a publisher’s website without users needing to click through strengthens The New York Times’ argument that OpenAI and Microsoft built AI products that can substitute its journalism, rather than use that content for a fundamentally different purpose, noted Danielle Coffey, CEO of News/Media Alliance. 

In legal terms, that “substitutive” use goes to the heart of fair use. “The apparent admission by the head of ChatGPT that their products are substitutive will be difficult to walk back – and likely eviscerates the fair use defense,” said Alan Chapell, privacy expert and founder of Chapell and Associates.

However, other legal experts say the documents, while damaging for OpenAI and Microsoft, fall short of a full game-changer. 

Substitution and market harm were already the most vulnerable parts of their fair-use defense, one lawyer told Digiday, speaking on the condition of anonymity. Most of the case materials remain sealed, and recent rulings in the Anthropic and Meta cases show courts can still view AI training as transformative use, the lawyer noted.

Even so, having senior execs at OpenAI and Microsoft describe AI products as “largely substitutive” and an “existential threat” to publishers, alongside references to paywall workarounds and “horse trading” over content, hands the Times fresh ammunition on those parts of the fair-use test. 

OpenAI’s paywall scraping puts it on wrong side of current federal law 

The evidence of bypassing paywalls and violating terms moves things out of the fair-use gray area and into “clearly stolen” territory: a very legal category, with much weaker fair-use arguments and much stronger claims for publishers, per Chapell. Going behind paywalls is a “really big deal” both legally and morally, he said. That’s OpenAI effectively admitting that its system substitutes for the publisher’s product, which “almost always” defeats a fair-use defense. 

Any evidence of deliberate paywall circumvention also puts OpenAI in potential violation of current federal law: the Digital Millennium Copyright Act – a 1998 U.S. federal law that updates copyright rules for the digital age. 

Open season on publisher scraping lawsuits with other AI companies

Like all court cases, this will be a slow burn, but it could open the door for more publishers to file lawsuits. If the Times wins – particularly around paywall circumvention and substitution – publishers get real bargaining power and more of them are likely to sue, which in turn pushes AI companies into a paid licensing market for certain categories of content. 

“This case will set precedent for all cases about training,” said a publisher exec at a major U.S. publication, who requested anonymity to speak freely. Google has trained Gemini, OpenAI has trained GPT, Anthropic has trained Claude, Meta has trained Llama. So all of these companies will say if this goes in favor of the New York Times, then there’ll be lots of suits.”

News/Media Alliance’s Coffey said that if the court determines that AI companies need to pay for content, every publisher will be either ready to file more lawsuits, or to come to the table to collectively license their content. “It’s already starting to happen,” she said. “It will come into full force once we have certainty of the law.”

Strengthens publishers’ terms of trade 

If the court buys the Times’ framing, it could force AI firms toward a paid market for news content. 

“Once that’s answered, I think it will be an across-the-board acknowledgement from the AI companies that they do need to come to the table,” Coffey said. “Publishers will collectively negotiate, and collectively license [their content]. I think it will be offered at scale, it will probably be usage-based, and the amounts paid will reflect the certainty of the law.”

For publishers, the unsealed material doesn’t bolster their legal case; it validates years of frustration about how their work has been treated.

“It’s nice to know that you know we weren’t being gaslit,” said a publishing exec at a major U.S. publication, who asked to speak anonymously, “but the people inside these companies know this too. Now we just need to work with these companies to get to the actual answer where this starts working.”

Spur’s Buttle argues that the big scraping cases (like NYT-OpenAI/Microsoft’s and ones like Reddit-Perplexity’s) are effectively a test of whether a real licensing market for AI will exist at all. If courts come down hard in favor of rights holders, they don’t just punish a few bad actors; they raise the price and risk of unlicensed scraping, which in turn forces AI developers toward properly paid, structured licensing deals with publsihers, rather than today’s mix of “backdoor routes” and risk-mitigation pay-offs. 

What to expect next

This is just the first wave of unredacted documents that publishing execs, lawyers and the public expect to see from this case. Legal teams for The New York Times, OpenAI and Microsoft will challenge each other on which documents should be unredacted and released to the public, and the judge will ultimately decide. That could take months. 

But many expect more court documents to be unsealed.

“This is only a taste of what’s to come. There’s a mountain of evidence to come, and the evidence weighs in our favor,” Coffey said.

Both sides of this case are asking Judge Sidney Stein for summary judgment, arguing that the evidence gathered during discovery is clear enough for him to resolve key issues in the case without sending them to a trial with a jury.