A summary-judgment motion from news-publisher plaintiffs led by The New York Times was unsealed on September 17, bringing internal Microsoft and OpenAI material into public view in the publishers’ copyright case against the two companies. The filing is not a ruling on infringement or fair use. But it adds documentary material to a dispute centered on whether AI systems may use journalism without permission and whether AI-generated answers harm the market for the underlying reporting.
The newly public material is presented through the publishers’ motion, rather than through a judicial finding. TechCrunch reported that the underlying exhibits cited in the brief remain sealed, limiting the public’s ability to assess the communications in their full context. That distinction is central: the filing contains the publishers’ characterization of the records and their argument that the records support claims of copying and market harm.
Among the most pointed passages is material attributed to Brent Hecht, Microsoft’s director of applied science. The publishers’ motion cited Hecht describing news scraping for AI training as “an astonishing theft of unprecedented proportions” and potentially “the largest theft of labor in human history.” The motion also cited internal discussions from OpenAI that, according to the reporting, described publishers as facing an “existential threat” and AI products as “largely substitutive.”
Referral traffic and the open web
The filing’s broader argument is that conversational AI and answer engines can reduce the need for users to visit a news site. According to Ars Technica and TechCrunch, the motion cited Microsoft data showing steep click-through-rate declines when users were given AI-generated answers instead of traditional Bing search results. TechCrunch reported a reduction of as much as 93% for The New York Times domain, while Ars reported differing ranges for different publisher plaintiffs. Those figures are evidence offered by the publishers in litigation, not a court-validated measurement applicable to all news queries or publishers.
The documents cited in the motion also portray Microsoft employees considering a feedback problem for the open web. A January 2024 Microsoft presentation reportedly called the possible cycle a “doom loop”: reduced traffic and revenue could weaken publishers and, in turn, diminish the supply of material used to support AI products. The filing quoted the presentation as describing news sources for large language models as a “content supply chain.” That framing matters because it links the publishers’ copyright claims to a practical question beyond model training: whether systems that summarize and answer from reporting can continue to rely on an economically viable journalism ecosystem.
Claims involving paywalled material
The publishers further alleged that OpenAI and Microsoft obtained or used news content through mass scraping, including material behind The New York Times paywall. Reporting on the unsealed motion described an exchange in which OpenAI researcher Nick Ryder referred to a paywall workaround and OpenAI President Greg Brockman replied, “Ah, nice.” The same reporting said the motion cited claims about publisher works appearing in training datasets. Because the cited exhibits are not public, those assertions remain allegations in the publishers’ brief rather than independently established findings.
The motion also cited testimony from Microsoft Chief Executive Satya Nadella. According to TechCrunch’s account, Nadella said that paywalled material should be licensed for grounding or training and that he would have used Microsoft’s contractual rights to require retraining had he known OpenAI had scraped and trained on paywalled information. The significance of that testimony, if characterized accurately in the motion, is that it bears on the parties’ factual disagreement over knowledge, data provenance and the handling of premium publisher content.
Microsoft’s response and the licensing alternative
Microsoft pushed back on the implications drawn from the internal material. In a statement to Ars Technica, the company said its AI products are transformative fair use and do not substitute for news sites. Microsoft also said Hecht’s comments represented an individual employee’s perspective, were not legal analysis and did not reflect the company’s views. Ars reported that OpenAI did not immediately respond to its request for comment; TechCrunch reported that neither company had returned its request at the time of publication.
The newly public filing arrives as licensing remains an alternative path alongside litigation. In July, Microsoft and Australian publisher Nine announced an agreement allowing Copilot to reference text from Nine’s mastheads beyond paywalled previews in AI searches, while presenting snippets, summaries and a route to the original publications. The arrangement does not resolve the U.S. publishers’ claims, but it illustrates the commercial model at stake: AI companies can seek authorized access to current, professionally produced reporting rather than rely on contested uses of material gathered from the web. (news.microsoft.com)
What the unsealing changes
For the litigation, the immediate consequence is evidentiary rather than dispositive. The motion gives the publisher plaintiffs publicly discussable material with which to argue that the defendants understood both the value of news content and the risk that AI answers could replace publisher visits. Microsoft and OpenAI can still contest the meaning, completeness and legal relevance of those records, as well as the publishers’ underlying copyright claims. No court, based on the materials described in the reports, has yet resolved whether the challenged uses constitute infringement or fair use.
The case therefore continues to test two competing accounts of generative AI’s relationship with journalism. Publishers contend that unlicensed copying and answer-generation appropriate reporting while undermining the audience and revenue needed to produce it. Technology companies have argued that AI uses are transformative and can help people find information. The unsealed filing does not settle that conflict, but it makes the fight over training data, referrals and licensing more concrete by placing parts of the companies’ own internal discussion at the center of the record.
