Social Rocket
Newly unsealed records in the copyright case between The New York Times, OpenAI and Microsoft are intensifying the debate over when using journalistic content for AI can still fall under fair use. The focus is not only on training data, but also on paywalls, substitution and existing licensing markets.
Digiday evaluated the latest records together with lawyers and publishing experts. The key point: the statements come from ongoing proceedings and are not yet a court finding of a legal violation.
Why paywalls become legally significant
The plaintiffs argue that OpenAI and Microsoft used copyrighted journalistic content on a large scale and, in some cases, also processed material behind paywalls. Newly unsealed internal communications and statements from executives are now being used to cast doubt on the fair use defense.
The question of whether technical protection measures were circumvented is particularly sensitive. Digiday cites legal experts who see a significantly different situation compared with freely accessible websites. An alleged circumvention of access restrictions could, in addition to classic copyright questions, also touch on other US legal provisions.
OpenAI disputes the plaintiffs' basic account. On 4September 2026, the company filed a motion for summary judgment and continues to argue that using content for model training may fall under fair use. On its official page on the New York Times case, OpenAI also emphasizes that parts of the lawsuits have already been narrowed or dismissed.
Substitution becomes the core issue for publishers
For publishers, the dispute is economically important because AI systems do not only process content but can deliver answers to users without them having to click through to the original website.
Reuters reported on 17September about internal statements in which employees and executives themselves discussed the risk that AI products could partly replace news offerings. According to Reuters, Microsoft said that individual internal statements do not reflect the company's legal position.
This puts the fourth fair use factor more sharply in focus: what economic impact does the use have on the market for the original work? If AI products reduce the need for a visit or subscription, that matters far more to publishers than the abstract question of whether a model technically "transforms" an article.
How strongly this zero-click effect is already affecting media business models is shown in our overview of Wikipedia between AI reach and declining direct visits.
Licensing markets already exist
Another point: the market for licensed publisher content is now real. Major AI and platform providers are signing agreements with media houses, while other publishers are focusing on direct machine-readable products.
Google, for example, is testing a pay-per-use compensation model for publisher content in AI products. Axios is building its own news feeds for AI agents and internal models with Axios Direct. Offers like these strengthen the argument that journalistic content has a concrete market value for AI systems.
For publishers, this changes the negotiating position: the more licensing becomes the established standard, the harder it becomes to treat unlicensed use as economically without consequence.
What the case means for marketing and publishing
For marketing teams, the case matters because content value increasingly arises beyond classic website visits. Brands, publishers and creators need to distinguish between visibility in AI answers, actual citations, referral traffic and commercial use of their content.
This is precisely why a single visibility metric is no longer enough for AI search. Our article on AEO measurement with Comscore data shows how much sourcing, citation and referral can differ.
For publishers, there is a second layer: when content serves as input for commercial AI products, licensing terms, technical access controls and documented usage rights become part of the monetization strategy.
Nothing is conclusively decided yet
The current state is not a ruling on whether OpenAI or Microsoft violated copyright or other regulations in this specific case. The unsealed documents give the plaintiffs new material; OpenAI and Microsoft maintain their fair use defense or dispute that individual internal statements define the companies' legal position.
For publishers, one trend is nonetheless clearly visible: the debate is shifting from "may models learn from public web material?" toward "what market value does content have, which access limits apply, and when does an AI product economically replace the original?"