On October 30, 2026, OpenAI published Fair use benefits everyone, a page of short excerpts from friend-of-the-court briefs filed in the copyright case brought by The New York Times and other news publishers. The message is blunt: authors, journalists, academics, librarians, accessibility advocates and even the United States government told the court that training AI on copyrighted works is fair use. OpenAI calls the Times lawsuit an attempt to "toss aside the longstanding principle of fair use" in "their own narrow pursuit of an undeserved payday."
That is advocacy, and it should be read as such. But the excerpts are real filings, and the arguments in them are the clearest preview of how the biggest AI copyright question of the decade may be argued. This post goes through who is speaking, the legal theories on offer, and what is conspicuously absent. For the Wikimedia side of the same case, see our earlier piece on the Wikimedia amicus brief and the training versus retrieval split.
TL;DR: what the page is and is not
| Question | Answer |
|---|---|
| Who published it? | OpenAI, on its own site, dated October 30, 2026 |
| What is on it? | Excerpts from amicus briefs by 11 named groups or people plus an "AI Progress" summary |
| Who wrote the briefs? | Third parties, including the US government, library and disability groups, professors, and creators |
| Who chose the excerpts? | OpenAI. The page says full filings are on the public docket |
| Is there a ruling? | None found. Summary judgment motions were reported filed in September 2026 |
| Does it change what you can build? | Not yet. It signals the arguments, not the outcome |
What OpenAI actually posted
The page opens with the company's framing: fair use "has played a critical role in driving innovation," it says, and experts across fields "have recognized that fair use is a win-win." It then lists the contributors, in its own order: American Council of the Blind with the Association of Research Libraries, American Library Association, Consumer Technology Association and Business Software Alliance as a joint group; Authors Alliance; Engine and Developers Alliance; Foundation for American Innovation; Joshua S. Gans; Kevin Kelly; Matthew Sag; Michael Wall and Sarah Rose Siskind; Public Knowledge; Ron Suskind and Brian L. Frye; and the United States Government.
Each gets a few sentences. The page links no filings directly in the text we read, but states that "the full filings are available on the public court docket." If you want to evaluate any claim here, that is where to go. The earlier, related announcement from the same company the same day, about letting sites state AI training preferences, is covered in our post on OpenAI honoring a shared robots.txt signal. The two pieces together show OpenAI pairing a legal defense of past training with a forward-looking opt-out for the future.
The arguments, one by one
Training is a different purpose
The unsigned "AI Progress" paragraph says AI training "uses copyrighted works for a fundamentally different purpose, does not provide the public with substitute access to those works, and does not cause the market harm cognizable under the Copyright Act." Legal expert Matthew Sag, a law professor, makes the technical version: under realistic conditions, model outputs "are in no relevant way similar to the invisible intermediate copies made during the training process." This is the transformative-use leg of the four-factor fair use test.
Market dilution is not market harm
Authors Alliance takes aim at a theory plaintiffs have used: that a flood of AI-generated works can reduce sales or visibility of older ones. The brief says new works can reduce the value of older works "without copying their protected expression," and copyright "deliberately leaves that competition outside the copyright owner's control." It asks the court to hold that "market dilution" is not cognizable harm under 17 U.S.C. section 107(4). This is the argument most likely to matter beyond the Times case, because it would also blunt claims by authors and musicians. We saw the same fight in the Anthropic book-destruction and Bartz fair use story and the music publishers' claims in Sony and Warner against Anthropic.
Search and library precedents
Public Knowledge points to older cases where courts protected tools that help people navigate information: Perfect 10 v. Amazon on image search, Authors Guild v. Google on Google Books, Authors Guild v. HathiTrust on accessible digital books, and ASTM v. Public.Resource.Org. Its conclusion: "Fair use will be equally beneficial in ensuring that AI's benefits accrue to the public at large, instead of being controlled by a privileged few." Whether a generative model is closer to a search index or a substitute for the source is exactly what the court must decide, and the page does not address the counterargument.
Startups and downstream users
Engine and Developers Alliance argues that "the purpose of training a foundation model is not to reproduce the works it learns; it is to build a general-purpose tool that others can adapt to new ends." The joint library, disability and tech-industry brief says restrictions would hit medical researchers, educators, people with disabilities, farmers and creators, many of whom "depend on systems developed by others, including open source AI systems." For builders using hosted or open models, this is the supply-side stake in the case.
Creators who use AI
Musician Michael Wall and writer Sarah Rose Siskind describe running their businesses with AI. Siskind is quoted as saying, "I don't think I could run my business without AI at this point." Author Kevin Kelly calls each model "an entirely new medium." These are personal testimonies, useful for color rather than legal weight.
Economics and national security
Economist Joshua Gans argues that copyright supports "the incentive to build tools that learn from that content," and that treating every displaced sale as actionable can weaken that incentive. The Foundation for American Innovation goes further, suggesting factor four should weigh benefits like new knowledge and even "the national-security value" of AI capabilities. That is a novel stretch of the doctrine and a point opponents will probably contest.
Journalists on the other side of the byline
Pulitzer-winning journalist Ron Suskind and law professor Brian Frye argue the news plaintiffs "retain and continue to exercise the exclusive right to sell consumers access to their articles," and that training did not impair it. Note the structure of this claim: it addresses the sale of access, not whether a chatbot reproduces articles, which is an issue the plaintiffs have raised elsewhere.
The US government's brief
The last excerpt is from the United States Government, which says the Times "seeks to narrow fair-use doctrine to exclude the training of OpenAI's large language models" and that this would be "inconsistent with basic copyright law principles." The filing was reported by outlets in early September 2026. Reporting summarized by AI Weekly says the Justice Department filed a statement of interest on September 1 backing OpenAI's defense, and that the Times plaintiffs later asked the court to give it "no weight at all." We could not independently open the docket, so treat dates and quotes beyond the OpenAI page as reported, not verified by us.
What the page leaves out
Selective excerpting means the page shows one side. Briefs from publishers, author groups and creators who oppose AI training are not here, and neither are the plaintiffs' core evidence or the court's questions. Several of the supporters are trade groups whose members build or sell AI products. The Consumer Technology Association and Business Software Alliance are industry bodies, and Engine represents startups. That is not disqualifying, but it is context.
The page also says nothing about the strongest plaintiff arguments, such as outputs that reproduce text, retrieval features that surface articles, or how training data was obtained. The Wikimedia brief we covered draws a line between training and retrieval, a distinction the OpenAI page blurs by speaking mostly about training.
Why this matters to people who build with AI
If courts keep treating training as fair use, the cost of building and fine-tuning models stays lower, and open-weight projects face less legal pressure. If they do not, expect more licensing deals and more emphasis on data provenance. Either way, the practical signals are moving in parallel: publishers want controls like the shared robots.txt token, and AI search is already squeezing them, as we explained in Google Zero and the publisher crawler-blocking debate. A running list of cases lives in our AI lawsuit tracker.
How to read a litigation PR page
A few habits help. First, check who picked the quotes. Second, find the filing on the docket and read the sentence before and after the excerpt. Third, separate claims about law (what fair use means) from claims about facts (what models do and what happened to the market). Fourth, remember that an amicus brief is persuasion for one side; courts weigh it lightly relative to the parties' evidence. Fifth, wait for the judge's reasoning, which may split issues such as pretraining, retrieval and output regurgitation.
What to watch next
- A summary judgment ruling on whether pretraining is fair use.
- Whether the court addresses market dilution as a legal theory.
- How OpenAI's promised opt-out signal is received by publishers.
- Any publisher-side briefs that OpenAI's page did not feature.
Bottom line
OpenAI's page is a well-assembled argument, not a verdict. It tells you which theories the AI side believes are strongest: transformative purpose, no cognizable dilution harm, public benefit, and precedent from search and library cases. It does not tell you how the judge will weigh them. Read the excerpts, then read the docket.
Details reflect public sources as of October 30, 2026, and the case is ongoing.