The Admission File: How a 92-Page Unsealing, a ‘Doom Loop’ Memo, and a Paywall Hack Put the Industry’s Own Words Into the Copyright Record

The Admission File: How a 92-Page Unsealing, a ‘Doom Loop’ Memo, and a Paywall Hack Put the Industry’s Own Words Into the Copyright Record

On September 17, 2026, Judge Sidney Stein unsealed the news plaintiffs’ combined summary judgment brief in the OpenAI copyright litigation — and the industry’s internal warnings about theft, substitution, and a “doom loop” for the web entered the public record.


What Was Unsealed, and Why It Matters

For nearly three years, the consolidated New York Times v. OpenAI and Microsoft copyright litigation has produced public filings in which both sides talked past each other: the publishers called the practice mass piracy, and the companies called it transformative fair use. On September 17, 2026, a 92-page combined summary judgment brief filed by the news plaintiffs — The New York Times, the New York Daily News, the Center for Investigative Reporting/The Intercept, and Ziff Davis — was unsealed in unredacted form, and for the first time the record contains what the defendants’ own employees said about the enterprise while building it [1]. The filing was assembled by Susman Godfrey and Rothwell Figg for the news organizations, but its most quotable lines are not the lawyers’ arguments; they are admissions drawn from internal documents and sworn depositions that Microsoft and OpenAI had fought to keep confidential [2].

The unsealing matters for two distinct reasons. First, the internal statements are in tension with the companies’ courtroom position that training is a transformative use which does not substitute for the works it learns from. Second — and less noticed — the documents were unsealed at all because the defendants’ own conduct during discovery created the pressure: plaintiffs’ counsel Steven Lieberman noted that the companies “insisted that these documents be treated as confidential so that the public could not see them,” and that “now the cat is out of the bag” [3].

The Admissions, In Their Own Words

The brief’s introduction frames the case around a single Microsoft document: this case is about, as Microsoft’s Director of Applied Science Brent Hecht put it, “an astonishing theft of unprecedented proportions,” perhaps the “largest theft of labor in human history” [1]. In a document composed soon after the suit was filed, Hecht explained that the systems “are useful because they replicate the patterns in their training content which, without ways of distributing economic value down the supply chain, necessarily threatens the economic stability of those who create the content” [1]. A separate Microsoft policy document compressed the problem into eight words: “LLMs are a product that destroys its supply chain.” [1]

The core logic of the plaintiffs’ evidentiary strategy is that a fair use defense lives or dies on substitution. Under the Supreme Court’s Andy Warhol Foundation v. Goldsmith decision, a use that substitutes for the market of the original work is “copyright’s bête noire,” and defendants cannot show a lack of market harm. So the brief does not primarily argue that the models memorized articles (though it documents verbatim output of Times and Daily News pieces); it assembles the defendants’ own statements that their products substitute for news:

  • The doom loop memo. An internal Microsoft document states: “Our AI content strategy has started a ‘doom loop’ that will hurt the performance of our models and the entire web at the same time: It is highly unusual that an end-product threatens the economic foundations of its essential suppliers, but that is the situation we have created for our LLM business with respect to its ‘content supply chain.’” [1]
  • The substitution quotes. OpenAI’s head of ChatGPT, Nick Turley, wrote that “[o]ur products are largely substitutive, period” and predicted they “will get more and more substitutive as they get better.” An OpenAI software engineer stated that “no matter how prominently we show the links, users won’t click.” [1]
  • The paywall hack. When OpenAI researcher Nick Ryder told co-founder Greg Brockman about “a hack to get around nytimes paywall” for scraping efforts, Brockman replied, “ah nice.” [1]
  • The traffic data. Microsoft’s own data recorded 83-93% drops in click-through rates for the Times and Daily News domains, and 51-94% drops for Ziff Davis properties, when Copilot answers replaced traditional Bing search results. [1]

Why do these admissions matter legally, beyond the bad optics? Fair use is a flexible, multi-factor defense, but after Warhol the market-effect factor is where AI cases are won or lost. The plaintiffs’ theory is that the defendants’ internal understanding — their own words showing they knew chatbots replace the underlying sources — is the most direct evidence available of market harm, and evidence that comes from the opposing party itself carries special weight at summary judgment [3]. The defense counters that an employee’s forward-looking warning is not a legal analysis of fair use, and that the proper test is whether outputs actually copy protected expression, not whether traffic patterns shift [4].

The Doom Loop, in Microsoft’s Own Numbers

The “doom loop” is not a rhetorical flourish; it is a description of a dependency. The filing quotes Microsoft data showing massive declines in click-through rates when Copilot answers a query instead of presenting search links. That dynamic is the core of the publishers’ market-harm case: if AI answers satisfy reader demand for news, the newsrooms that produced the answers lose the readers, subscriptions, and advertising that fund original reporting — and the models that learned from that reporting eventually face a thinner, degraded supply of high-quality content to learn from.

It is also the logic of a collective action problem, and the brief makes that argument explicitly. As the news groups put it, “AI companies remain powerless to break out of this ‘doom loop,’ because, while the industry as a whole would benefit if every company paid to sustain the continued production of the creative works their technology depends on, each individual company is better off taking content for free while others pay.” [3] The reasoning is a classic free-rider trap: licensing is costly to any single firm, but free-riding is only rational until the suppliers collapse — a dynamic that harms both the web and the AI industry that depends on it.

The Defense and the Distancing

Microsoft has moved to neutralize the admissions by personalizing them. A spokesperson said Hecht’s memos “reflect one employee’s individual perspective, are not a legal analysis, and do not represent the company’s views,” and a Microsoft filing described Hecht’s role as bringing “asymmetrical, futuristic, and academic points of view” to the company [5]. The company characterized CEO Satya Nadella’s sworn testimony — that chatbots have “substituted” for going to “the underlying source” — as broad “observations” about how people consume information, “not conclusions about copyright questions before the Court” [3].

The distinction the companies draw is between describing a market trend and conceding a legal conclusion. That separation is genuine as a matter of law — a fair use ruling does not turn on whether an executive thought the scraping was “fair” — but the filings give Judge Stein both kinds of material, and the plaintiffs want the admissions read as evidence of intent and of anticipated market effect, not as stray opinion [4]. Expect the companies’ oppositions to argue the memos are inadmissible hearsay or irrelevant to the statutory factors; expect the plaintiffs to respond that intent and purpose under factor one of the fair use analysis is exactly what these documents establish.

What Happens Next

The unsealed brief is one side of the summary judgment record, not a ruling. The case, In re OpenAI, Inc., Copyright Infringement Litigation (No. 1:25-md-03143, S.D.N.Y.), consolidates the Times’ suit with the Daily News, Intercept, and Ziff Davis complaints, and both sides filed competing summary judgment motions in early September [4]. Microsoft and OpenAI have not yet filed their full response to the newly unsealed exhibits, and the companies’ reply briefs are the next major filings in the case [5]. Judge Stein could rule on fair use as a matter of law, send the case to a jury on specific infringement claims, or split the difference — for example, granting summary judgment on the plaintiffs’ DMCA claims (which turn on removal of copyright management information) while reserving fair use for trial.

Three things make this filing consequential beyond one case:

  • It resets licensing negotiations. Every publisher currently negotiating with AI companies now has a public record in which the buyers’ own employees describe the extraction as theft and quantify the traffic collapse it causes [2].
  • It supplies the collective action argument. The doom loop framing gives policymakers a concrete account of why voluntary licensing fails — no single firm has an incentive to pay while its competitors free-ride — which is the standard predicate for a compulsory licensing or remuneration regime [3].
  • It is the first case where AI firms’ internal knowledge is in the record. Whether or not Judge Stein relies on it, future plaintiffs will demand the same category of discovery: what did the company know, and when [4].

References

[1] U.S. District Court, S.D.N.Y. — [PRIMARY] “News Plaintiffs’ Combined Summary Judgment Brief.” In re OpenAI, Inc., Copyright Infringement Litigation, No. 1:25-md-03143-SHS-OTW, ECF No. 1977-1. September 17, 2026.

[2] 404 Media — [SECONDARY] “‘Doom Loop’: OpenAI and Microsoft Admits LLMs Are Destroying the Web and Built on Theft” by Jason Koebler. September 17, 2026.

[3] Ars Technica — [SECONDARY] “Microsoft exec called AI scraping the ‘largest theft of labor in human history’” by Ashley Belanger. September 18, 2026.

[4] Reuters — [SECONDARY] “OpenAI, New York Times case tees up key test of AI training under copyright law” by Blake Brittain. September 8, 2026.

[5] AI Chat Daily — [SECONDARY] “Unsealed NYT filing: OpenAI and Microsoft knew AI would trigger a web ‘doom loop’” by Jaeden Schafer. September 18, 2026.


AI automated and generated content. Human reviewed.

The Weekly AI Policy Brief

AI policy moves fast. Get the week’s developments, new trackers, and podcast episodes in one short email — free, ad-free, no spam.

Comments

Leave a Reply

Your email address will not be published. Required fields are marked *