OpenAI’s “Fair Use Benefits Everyone” Page Turns a Docket Into a Highlight Reel: None of Its 20 Excerpts Backs the Plaintiffs, at Least 11 Filers Do, and the Court Dismissed One Filer’s Motion the Day It Ran

On October 30, OpenAI published a page titled “Fair use benefits everyone.” It excerpts twenty filings from the multidistrict copyright case in Manhattan, where the New York Times, other news publishers and a class of book authors are asking Judge Sidney H. Stein to reject fair use as a defense for training its AI models. The page says experts and leaders “have shared their perspectives with the Court in filing briefs in support of fair use,” and it adds that “the full filings are available on the public court docket.”

The page is advocacy by a party, and it reads that way. I wanted to know what it leaves out, so I read it against the docket: the court’s order on amicus briefs, the order that moved the deadline, the Justice Department’s filing, the plaintiffs’ response to it, the docket sheet of the Times case through October 30, and about 150 filings from the free CourtListener archive. Everything below is as of November 1, 2026.

The short version

1. None of the twenty excerpts supports the plaintiffs, but at least 11 filers say they do. I identified 38 amicus filers on the docket. Of the 24 whose own papers state a side, 11 back the plaintiffs, 9 back the defendants and 4 say they back neither party. The page excerpts 8 of the 9 defense filers, 2 of the 4 neutral ones and none of the 11 plaintiff-side filers (Table 2).
2. The excerpts come from proposed briefs. Nineteen of the twenty entries are amicus filers, and the court’s August 7 order made each of them ask for leave. The docket sheet I read, which runs through October 30, shows no order granting leave to anyone. The page does not mention the requirement and links no filing: both links in its body go to OpenAI’s own New York Times page.
3. The court dismissed one filer’s motion the day the page ran. On October 30 Judge Stein dismissed Brian Frye’s letter motion because it did not request leave. The brief Frye filed with Ron Suskind is captioned “in Support of Neither Party” and discloses that OpenAI owns three percent of Suskind’s AI company. The page calls Suskind a “Pulitzer Prize-winning journalist” and mentions neither point.
4. The twentieth entry is not an amicus brief. “United States Government” is the Justice Department’s September 1 statement of interest under 28 U.S.C. § 517, which, the filing notes, “does not require the Court’s leave.” On September 28 the News Plaintiffs asked the court to “give the DOJ’s statement no weight at all.” The page does not say so.
5. The excerpts are accurate where I could check them. Nine of the twenty matched their filings word for word, apart from PDF artifacts. I could not retrieve the other eleven.
6. The page never explains the theory two of its excerpts attack by name. Authors Alliance’s whole brief asks one question, whether “market dilution” is a cognizable harm. The idea comes from a 2025 ruling in Kadrey v. Meta that both sets of plaintiffs cite and the government’s filing criticizes. The page mentions neither the ruling nor the Copyright Office report the plaintiffs set against the government.

None of this makes the page false. It makes it a party’s selection, and the rest of this article shows what the selection omits and what each omitted fact rests on.

What OpenAI published

The page opens with a claim about who is on its side: “Authors, journalists, creators, academics, accessibility advocates, legal experts, librarians, economists and business leaders are voicing their support for the longstanding principle of fair use and its application to AI training.” It describes the Times’ lawsuit as asking “everyone to ignore a half century of legal history and progress,” in a “narrow pursuit of an undeserved payday.” Then it lists twenty names and gives each one a paragraph “in their own words.”

Two things about the page itself are easy to verify. It carries no link to any filing; the only two links in its body, on “The New York Times’ lawsuit” and “here,” both point to OpenAI’s own /new-york-times/ page. And it was published the same day as OpenAI’s robots.txt training-preference post, which we analyzed on October 30. That is my observation, not a claim by OpenAI, but the two posts bear on one question, who gets to say no to AI training, one aimed at crawlers and one at the court.

The case behind the page

Where it stands

The case is In re OpenAI, Inc., Copyright Infringement Litigation, No. 1:25-md-03143 in the Southern District of New York. The Judicial Panel on Multidistrict Litigation centralized the suits on April 3, 2025, according to a stipulation in the docket (ECF 2072). The News Plaintiffs (the Times, the Daily News, the Intercept, Ziff Davis and others) and the Class Plaintiffs (authors) have moved for summary judgment, OpenAI and Microsoft have cross-moved, and fair use is the central question. A September 24 order, which endorsed the plaintiffs’ letter (ECF 2049), set this schedule:

I found no hearing date and no ruling on the docket.

The ruling will reach past the lead cases. At least four later suits are parked behind it. Sullivan et al. v. OpenAI Foundation was stayed by order on September 16, “pending resolution of” the summary judgment motions in the other active cases (ECF 1967). The Seattle Times and Newsday, whose complaint we covered in September, filed a stay stipulation on September 28 (ECF 2072). Times Publishing Co. and 26 co-plaintiffs in Times Publishing Co. et al. v. Microsoft filed one on October 14 and again on October 15 as a stipulation and order (ECF 2104), and USA Today Co. and its co-plaintiffs filed theirs on October 16 (ECF 2110). The Seattle Times stipulation also says the parties will file a joint status report “within 30 days of rulings” on the motions.

The rule every amicus had to follow

On August 7 the court ordered (ECF 1663): “Anyone wishing to file an amicus brief in connection with the summary judgment motions in this multidistrict litigation shall file a motion requesting leave to file an amicus brief no later than October 16, 2026. The motion requesting such leave shall include the proposed amicus brief, which shall be no more than 6,000 words.”

The plaintiffs asked to move the amicus deadline in step with the oppositions, “so that amici could review both sides’ opening and opposition briefs” (ECF 2043), and the court granted the schedule above. The new deadline explains why the docket holds almost no amicus filings until the last week. The AI Rights Project filed on October 14 and Kevin Kelly on October 16; nearly everyone else filed between October 28 and 30.

Until leave is granted, a filing like this is a proposed brief. One movant quotes United States v. Ahmed for the point that “district courts have broad discretion to permit or deny the appearance of amici curiae in a given case” (ECF 2306). In the Times docket sheet, which I have as an unbroken run of entries 1626 to 1831 (September 25 to October 30), I found one order about an amicus motion, the dismissal of Frye’s, and none granting leave. The page says the filers “have shared their perspectives with the Court in filing briefs.” That is true of lodging them. Whether the court will read them is undecided as far as the docket shows.

The twenty entries, scored

For each name on the page I looked for the filer’s motion or brief on the docket and recorded the side its own caption, motion text or docket text states. I did not infer sides from reputation. “Not stated” means I could read the entry but its text does not say which side it supports.

Read down the third column. Eight entries state support for the defendants, two state support for neither party, and the government’s statement is its own kind of filing. Nine I could not classify: six I found on the docket without a stated side, and three I could not find at all. No entry states support for the plaintiffs.

The 11 filers who are not on the page

Table 2 lists every filer I found whose own papers say it supports the plaintiffs. The page’s opening sentence credits fair use to “authors, journalists, creators, academics” and “economists.” The plaintiff side has those too: performers’ and writers’ unions (SAG-AFTRA, the American Federation of Musicians and the two Writers Guild chapters), a newsroom (ProPublica) and news-industry trade groups, eleven First Amendment scholars, and Professor Joseph Stiglitz, whose motion says he received the 2001 Nobel Memorial Prize in Economics.

Two caveats. The AI Rights Project is the narrowest: its motion says it takes “no position on the parties’ factual disputes” and addresses two questions the parties’ briefs do not raise: whether the creative judgment behind transformative use must be human, and how far a court may reshape the statute’s balance. And these are captions, not assessments of the briefs’ merits.

The page also omits two filers who say they support neither party, the Wikimedia Foundation (ECF 2361) and the technology investor Michael Moritz (ECF 2434); one more defense-side filer, the Copia Institute (ECF 2369); and eight filers whose papers I could read but whose side they do not state: Samuelson, Sprigman and Schultz; the Center for American Rights and its co-movants; the Human Consent Foundation, Creative Artists Agency and their co-movants, who include Cate Blanchett and Steven Soderbergh; Progressive Democrats of America and its co-movants; Members of Congress; America’s Newspapers; the Institute for Intellectual Property and Social Justice; and William Morris Endeavor. In all, the page covers 16 of the 38 filers I identified and omits 22.

Three entries that need more than an excerpt

Suskind and Frye: a dismissed motion, a disclosed stake, and a concession the excerpt skips

The brief was filed on October 28 (ECF 2283). Its caption reads “in Support of Neither Party.” Its identity-and-interests section says Ron Suskind is the CEO of Sourcebase, an AI company, and that “the company recently received a minority investment amounting to three percent ownership from OpenAI. OAI is a passive investor and has not seen or influenced this brief in any way, nor has the New York Times.” That disclosure belongs in the brief, and it is there. It is not on OpenAI’s page, which labels the pair “Pulitzer Prize-winning journalist / Professor of Law.”

On October 30 the court dismissed the motion (ECF 2412): “the filing does not request leave to file the amicus brief, in violation of the Court’s Order on amicus filings. … Mr. Frye may refile if he chooses.” The order lets Frye refile, so the brief may come back. The page, dated the same day, lists it among the submissions “in support of fair use” and says nothing of the dismissal; I could not tell which came first.

The excerpt is accurate, and it sits a few paragraphs after a passage the page does not carry. The brief says the News Plaintiffs’ harm claims are “superficially true”: OpenAI’s use “did harm the market for and value of those articles,” it “did discourage licensing newspaper articles for AI training by showing it isn’t required,” and “it did dilute the market for newspaper articles by creating a lot more of them.” Then it argues that “none of those harms are relevant, because they were all the result of fair competition.” The sentence OpenAI excerpts, that the use “didn’t harm the reasonable commercial expectations of the News Plaintiffs in any way,” is the brief’s conclusion on one specific question. The brief also proposes a middle path, saying the court can serve every interest by “requiring generative AI models to attribute their outputs to the newspaper articles on which the models were trained.” Its conclusion does lean toward OpenAI’s result, warning that if training is not a fair use, “a few large copyright owners will receive an immense windfall.”

The Justice Department: a statement the court did not have to admit

The statement of interest is ECF 1682, 20 pages, filed September 1. It begins: “The United States appears pursuant to 28 U.S.C. § 517 to inform the Court of its interest in this litigation.” A footnote, quoting a district court, says the statute “contains no time limitation and does not require the Court’s leave.” That makes it the only one of the page’s twenty entries not governed by the August 7 order, and the only one that needed no permission to be on the docket.

The page excerpts the filing’s opening section accurately. Its ask is framed in general terms: “the United States has a strong interest in this Court rejecting any argument that training LLMs on copyrighted texts violates copyright law.” The closing paragraph also concedes that “the fair-use inquiry hinges on the specific facts and uses at issue in each case,” then calls it “problematic” and “legally incorrect” to impose liability that would generally make training impermissible without licensing. One footnote says the government “takes no position on whether a licensing regime would be financially or logistically feasible,” and another says its arguments “apply similarly to all parties in this litigation and the related cases,” which is why a statement that names only the Times and OpenAI also speaks to the authors’ case.

The News Plaintiffs answered on September 28 (ECF 2071). They say the statement came “just a few days before the parties filed their motions for summary judgment,” so the government wrote it “without the benefit of seeing the parties’ evidence or briefing”; that it rests on the administration’s pro-AI policy rather than copyright doctrine; that it clashes with the Copyright Office’s May 2025 report on generative AI training; and that “the DOJ reportedly did not contact any relevant agency,” a point they source to an Axios scoop that I could not open, because it sits behind a bot check. The government’s own footnote 17 attacks the Register of Copyrights’ “threadbare reasoning.” The plaintiffs conclude that “the Court should give the DOJ’s statement no weight at all.” OpenAI’s page quotes the statement and does not mention the response.

Market dilution: a theory two excerpts attack by name

Authors Alliance told the court that its brief “addresses a single legal question”: whether “market dilution,” meaning competition from non-infringing works that a model trained on a plaintiff’s works can generate, “is a cognizable harm under the fourth fair use factor” (ECF 2328). The page’s excerpt from Ed Lee and Rebecca Tushnet calls the plaintiffs’ theory of “dilution” “unprecedented” and says it “turns copyright on its head.” The government’s filing says the court in Kadrey “did not cite any case supporting such a theory” (footnote 17) and illustrates its objection with Joan Didion typing out Hemingway’s stories to learn how sentences worked.

The filings show where the theory came from. The Class Plaintiffs call the “harm of market dilution” their first market harm, citing Kadrey, 788 F. Supp. 3d 1026, 1052 (ECF 1982). The News Plaintiffs’ combined brief quotes the same ruling: “it seems likely that market dilution will often cause plaintiffs to decisively win the fourth factor” (ECF 2108). OpenAI’s own memorandum has a heading calling the theory “not legally cognizable” (ECF 1983). A reader of the page sees the attacks and never learns that the theory is a federal judge’s reasoning that both groups of plaintiffs are relying on. Neither Kadrey nor the Copyright Office’s report appears anywhere on the page, though the government’s filing and the News Plaintiffs’ papers discuss both, and the Class Plaintiffs rely on Kadrey.

One more excerpt deserves a look in this light. The Foundation for American Innovation, which filed in support of OpenAI on the fourth factor, says its argument “supports a novel application here: considering the national-security value of capabilities that enable new analysis, software, and other productive work” (ECF 2378). The page’s opening sentence speaks of a “longstanding principle of fair use.” Those do not conflict, since a principle can be old while an application is new, but the page includes a filer who calls its own argument novel and does not flag it.

Are the excerpts accurate?

I split each excerpt into sentences and searched the filing text for each one after normalizing quotes and whitespace. Nine matched: A.I. Progress, Authors Alliance, Engine and Developers Alliance, the Foundation for American Innovation, Kevin Kelly, Matthew Sag, Public Knowledge, Suskind and Frye, and the United States. The sentences that did not match at first all traced to page numbers or footnotes interleaved mid-sentence by the PDF, or to a line-break hyphen. I found no case where the page changed a filer’s meaning. I could not check the other eleven: three I could not find on the docket, and eight whose briefs I did not retrieve from the archive.

What to watch next

• The leave rulings. Look for an order granting or denying the October 28 to 30 motions, and for whether Frye refiles.
• November 20. Replies are due, and they will show whether the defendants engage with the plaintiff-side amici the page does not mention.
• The statement of interest. The plaintiffs asked for no weight; the docket I read has no ruling on that.
• The stays. At least four later cases wait on the summary judgment ruling, and the Seattle Times parties have promised a status report within 30 days of it.
• How any ruling treats dilution. Two of the page’s excerpts name it, the government’s filing criticizes it, and both sets of plaintiffs rely on it.

A short checklist for reading a party’s consensus page

1. Does it link the filings, or only the party’s own pages?
2. Are the submissions accepted or merely lodged? Find the court’s rule on leave.
3. Search the docket for the kind of filing the page describes, here amicus motions, and read the captions on the other side.
4. Read each filer’s interests section for ties to the party.
5. Read the paragraphs before an excerpt; concessions tend to sit there.

How I checked, and what I could not see

I used two sources. The first is the CourtListener docket sheet for the Times case (No. 1:23-cv-11195), where amicus motions filed in the multidistrict case are cross-posted; I have entries 1626 to 1831. The second is the free RECAP archive of the multidistrict docket, from which I read the filings I cite. I did not use PACER, and I could not load the multidistrict docket sheet itself, so my filer counts are a floor: a filer that appears only on the multidistrict docket, as the AI Rights Project does, is in my count only because its papers were in the archive. Entry numbers in this article are multidistrict docket numbers unless I say “Times docket.”

The archive’s file names follow a pattern, so anyone can pull the same documents. The number after the case ID is the entry, and the final digit is the attachment (0 is the main document, 1 is the first attachment, such as a proposed brief):

curl -sL -A "Mozilla/5.0" -o 1682.pdf \
  "https://storage.courtlistener.com/recap/gov.uscourts.nysd.640396/gov.uscourts.nysd.640396.1682.0.pdf"
pdftotext -layout 1682.pdf 1682.txt

The command assumes curl and the poppler pdftotext tool. Entry 1682 is the Justice Department’s statement. The archive rate-limits bursts, so space the requests out.

Sources

• OpenAI, Fair use benefits everyone (October 30, 2026), read as served on November 1.
• Court orders: ECF 1663 (August 7, amicus procedure), ECF 2049 (September 24, schedule), ECF 2412 (October 30, Frye), ECF 1967 (Sullivan stay).
• Statement of Interest of the United States, ECF 1682 (September 1) and News Plaintiffs’ Response, ECF 2071 (September 28).
• Summary judgment papers: Class Plaintiffs, ECF 1982, OpenAI, ECF 1983 and News Plaintiffs, ECF 2108.
• Amicus filings are linked in Tables 1 and 2 and in the text. The Times docket sheet is on CourtListener.

Related coverage on sxz.io

• OpenAI’s Shared Training Signal Turns robots.txt Into a Purpose Token From a Same-Day Individual Draft, the other OpenAI post of October 30.
• Seattle Times and Newsday Sue OpenAI and Microsoft Over a ‘Snake Eating Its Own Tail’, one of the four later suits now stayed behind this ruling.
• OpenAI’s Usage Policy Rewrite Turns Model Gateways Into an Approval Request and Fences Off Training Signals, another revision-to-revision audit of an OpenAI policy document.