Friday was the deadline for amicus briefs in the copyright fight over ChatGPT, and the filings split along the line that matters most. OpenAI’s allies argue that training AI on copyrighted text is fair use. Publishers, unions and news groups say it is not, and they want a federal judge in Manhattan to reject that defense.

These OpenAI fair use amicus briefs were filed in In re OpenAI Copyright Infringement Litigation, a consolidated group of 12 lawsuits before Judge Sidney Stein. The New York Times’ 2023 suit against OpenAI and Microsoft is part of that group, according to TheNextWeb’s coverage of the Wikimedia brief. Whatever Stein decides on fair use would be the first large U.S. ruling on whether training generative AI on copyrighted text is fair use. That question touches every AI company and every publisher, author and creator who licenses content.

What the OpenAI fair use amicus briefs set in motion

The case has moved through a fixed sequence. Cross-motions for summary judgment on fair use were filed September 4, 2026, and opposition briefs were due October 23. Today’s deadline added outside voices to the record, and the filings are now public before summary judgment briefing closes.

Stein has not set a hearing date or a ruling schedule. He has not yet ruled on fair use, though he rejected OpenAI’s bid to dismiss in March 2025.

The reply deadline is unsettled. One tracker lists November 6, while two others list November 20. Check the court docket before relying on either date.

Where OpenAI’s allies stand

OpenAI published a roundup on October 30 quoting briefs from filers that back its position that training AI on copyrighted text is fair use. According to secondhand accounts of that post, the filers include creators, academics, accessibility advocates, librarians, economists and business leaders. The U.S. government is among them, and twelve filers are quoted in the roundup.

The roundup also goes after the Times. It describes the newspaper’s request as an “undeserved payday.” That phrasing suggests OpenAI wants the debate framed as access to knowledge rather than as a licensing dispute.

This article relies on secondary reports of OpenAI’s post, because the post itself was not reviewed directly for this story. The full list of filers and the contents of the government’s brief were not verified beyond OpenAI’s own summaries, as Future Tools’ news roundup also notes. Treat the filer count as OpenAI’s description, not an independent audit.

The government’s position is also on record in a separate filing. In early September, the Justice Department filed a statement of interest saying that barring AI training on copyrighted material could severely hamper progress and that training is not in itself a copyright violation. That position fits the side OpenAI is now promoting.

Publishers, unions and Wikimedia push back

News groups, publishers and Hollywood unions filed briefs asking Stein to reject OpenAI’s fair-use defense. One brief from the Human Consent Foundation, backed by Cate Blanchett and joined by CAA and other groups, supports the Times. It was filed late Friday in the Southern District of New York, as reported by Mezha.

A separate brief from SAG-AFTRA, the Writers Guild and the American Federation of Musicians makes a labor argument. Accepting the companies’ fair-use claim, the unions say, would make negotiated union contract terms moot. That turns a copyright dispute into a question about whether bargained protections for AI use mean anything. For performers, writers and musicians, the answer depends on whether courts treat training as free use.

Wikimedia’s filing offers the most specific proposal in the set. The foundation filed a neutral amicus brief. It says training is likely fair use, but that chatbots using retrieval to copy and answer from Wikipedia text are not, and it asks the court to split the question in two. Wikimedia’s own explainer lays out the reasoning. TheNextWeb quotes the brief saying OpenAI’s use of copies in ChatGPT’s RAG feature, short for retrieval-augmented generation, “is not fair.”

Not every filing comes from the creator side. A brief from the Center for American Rights, one of several conservative groups in the filings, warns that an expansive fair-use ruling would let foreign adversaries off the hook. It says accepting OpenAI’s view would “encourage foreign adversaries” to continue “this obvious form of IP theft.” That argument moves the case toward national security, a framing that could resonate in Washington, where AI rules have become a midterm campaign issue.

Training versus output is the split that matters

The sharpest disagreement in the filings is not whether AI learns from text. It is whether copying text to train a model differs from copying text to produce answers. Wikimedia’s brief and the Justice Department’s filing both separate training from chatbot output.

That separation carries practical weight. Training copies works into a dataset so a model can learn patterns. Retrieval features fetch live pages and reproduce or summarize them in response to a user’s question. A judge could find that training is fair use while retrieval is not, which would leave AI companies with different legal rules for different products.

The case centers on training and on retrieval features, so the outcome could affect how assistants that pull live web content are built and priced. Businesses that rely on ChatGPT or similar tools would likely feel that first, through product changes and pricing decisions rather than through any order that applies to them directly.

For publishers and newsrooms, the line decides whether licensing fees become legally necessary for training, for retrieval or for both. The case summary from Ferrando’s coverage of the Third Circuit ruling ties this to the licensing market that court recognized. Freelance writers and small outlets have the least bargaining power, so they would feel any new rule most sharply.

How the Third Circuit and the Anthropic settlement set the stakes

The first U.S. appeals ruling on AI training came from the Third Circuit on September 29, 2026, in Thomson Reuters v. Ross. The court held that copying Westlaw headnotes to train a rival legal search tool was not fair use. It left open whether generative AI training qualifies. That gap is why a district-court ruling from Stein would carry so much weight, since it could become the first large decision on generative AI itself.

Another reference point comes from the Anthropic authors’ case. According to a Norton Rose Fulbright update on AI copyright cases in 2026, a court found that AI training on copyrighted books was fair use, but that storing pirated copies was not. The case then settled for $1.5 billion, roughly $3,000 per work. That figure is likely to become a reference point in licensing talks, even though it reflects a dispute over pirated copies rather than a fair-use verdict on training.

The OpenAI litigation is also a coordinated proceeding. Other news publishers’ suits against OpenAI are handled in the same proceeding, so a ruling from Stein could shape many cases at once rather than one newspaper’s claim.

Pressure on AI companies also shows up outside court. Reddit has blamed Google’s AI search for a traffic drop and plans a login wall, which signals that platforms are moving to control access to their content.

When a ruling could come and what to watch

No ruling date exists yet. The next checkpoints are the reply briefs, whether Stein sets a hearing, and whether he separates training from output as Wikimedia asks. Each of those would tell readers more than any tracker’s summary.

A ruling would likely be appealed by the losing side. Appeals from federal courts in Manhattan go to the Second Circuit, which would be an early appellate voice on generative AI training if the question reaches it. The Third Circuit’s narrow reading leaves that broader question for later courts.

What creators and rights holders can do now

No action is required from general readers. Publishers, freelancers and rights managers have a few concrete steps worth taking while the briefing runs:

  • Pull the docket for In re OpenAI Copyright Infringement Litigation in the Southern District of New York and confirm the November reply deadline, since trackers disagree.
  • Read your licensing agreements to see whether they treat AI training, retrieval and display differently. The Stein case draws that line, so contract language should address it too.
  • If you belong to a guild or union, check your contract’s AI terms. The SAG-AFTRA, Writers Guild and musicians’ brief argues those protections depend on how the court rules.
  • Keep records of where and how your content appears in AI tools, since any future damages or licensing claim will depend on evidence.

The next concrete milestones are the reply briefs and whatever Stein does with them. If he answers the training question and the output question separately, AI companies could face different rules for building models and for answering questions with live content. Either way, the first large ruling on training will set terms for publishers, authors and AI companies negotiating deals now.