Stack of hardcover books beside a laptop displaying AI code and the OpenAI logo, with a judge's gavel and copyright document illustrating the OpenAI copyright lawsuit over authors' works used in AI training.

OpenAI Copyright Lawsuits: Authors vs. AI Training Data, and Where the Fight Stands in Late 2026

Bywpusername6782

Published October 28, 2026

The authors’ copyright case against OpenAI and Microsoft has reached the stage that will decide it. On September 4, 2026, the Authors Guild and a group of best-selling novelists asked a federal judge in Manhattan to rule, without a trial, that the companies infringed their copyrights by copying books, many allegedly taken from pirate websites, to train the models behind ChatGPT. OpenAI and Microsoft filed their own motions the same day asking the court to find that training is protected fair use. Opposition briefs were filed October 23, and reply briefs are due November 20.

The outcome will be one of the most closely watched copyright decisions in a generation. It sits inside a wider legal map that changed sharply this year: a record $1.5 billion settlement between authors and Anthropic won final approval in July, a federal appeals court issued the first appellate ruling on AI training and fair use in September, and the U.S. Justice Department formally sided with OpenAI. This report explains what the authors claim, how OpenAI defends itself, how courts have handled similar cases, and what comes next. All claims against OpenAI and Microsoft described here are allegations; no court has decided them.

Where the OpenAI Copyright Case Stands

The Authors Guild and 17 fiction writers first sued OpenAI on September 19, 2023, and added Microsoft, OpenAI’s largest investor and cloud partner, later that year. Other author suits followed in New York and California, including a nonfiction writers’ case and the earlier California cases brought by Paul Tremblay, Sarah Silverman and others. On April 3, 2025, the Judicial Panel on Multidistrict Litigation centralized the author cases with The New York Times‘ lawsuit and other news publisher suits before Judge Stein. Inside the MDL, the author claims proceed as a single consolidated class complaint.

Named plaintiffs in the consolidated author case include David Baldacci, Taylor Branch, Michael Connelly, Sylvia Day, Jonathan Franzen, Christopher Golden, Andrew Sean Greer, John Grisham, David Henry Hwang, George R.R. Martin, Jodi Picoult, Stacy Schiff and James Shapiro, along with the Authors Guild. Fact and expert discovery closed in the summer of 2026.

DateEvent
Sept. 19, 2023Authors Guild and 17 novelists file proposed class action against OpenAI in S.D.N.Y.
Dec. 2023Microsoft added as a defendant; The New York Times files its own suit.
Apr. 3, 2025Author and news cases centralized as MDL No. 3143 before Judge Stein.
Oct. 27, 2025Judge Stein denies OpenAI’s motion to dismiss the authors’ claim that ChatGPT outputs infringe.
Feb. 6, 2026Judge Stein sets aside an order that would have forced OpenAI to disclose privileged communications about deleting two book datasets.
Sept. 1, 2026U.S. Justice Department files a statement of interest backing OpenAI’s fair use position on training.
Sept. 4, 2026Authors, news publishers, OpenAI and Microsoft file cross-motions for summary judgment.
Sept. 17, 2026Redacted versions of the plaintiffs’ briefs, quoting internal company documents, are made public.
Oct. 23, 2026Opposition briefs filed.
Nov. 20, 2026Reply briefs due. No hearing or trial date set.

Sources: Authors Guild; court docket for MDL No. 3143 as reported by OpenClassActions, Sterne Kessler and others. Schedule dates reflect the court’s September 24, 2026 amended order as reported.

What the Authors Allege

The authors’ summary judgment brief asks the court to hold OpenAI and Microsoft liable for copying their books at several points, and to reject fair use at each one.

Acquiring books from pirate libraries

The plaintiffs allege that OpenAI built two early training datasets, known internally as Books1 and Books2, from material downloaded from Library Genesis (LibGen), a shadow library that courts have repeatedly ordered shut down. According to the authors’ unsealed filings, OpenAI deleted its LibGen files in 2022. OpenAI has said the datasets were removed because they were no longer used. In a filing that became public in September 2026, the authors also allege that Microsoft learned of OpenAI’s use of LibGen in 2019, when an early version of GPT-3 was presented to company leadership.

Training the models

The core infringement theory is that copying entire books into training datasets and feeding them through the training process reproduces protected expression without a license. The authors argue that the purpose is commercial and that the resulting models compete with human writers in the market for books.

Outputs that resemble the books

The authors also claim that ChatGPT can generate detailed summaries, outlines and sequel-style text drawn from their novels. In October 2025, Judge Stein refused to dismiss that output-based claim. Using ChatGPT summaries of George R.R. Martin’s A Game of Thrones as an example, he found that a reasonable jury could conclude some outputs are substantially similar to the plaintiffs’ books because they reproduce protected elements such as plot, characters and setting. He expressly did not decide whether those outputs are fair use.

Market harm

The authors describe harm to the book market as the most important fair use factor. Their brief argues that OpenAI’s models pose an existential threat to authors and publishers, pointing to a growing volume of AI-generated books and to internal statements by OpenAI personnel about AI substituting for human writing. Those characterizations come from the plaintiffs’ brief and are disputed.

How OpenAI and Microsoft Defend Themselves

OpenAI’s position is that training large language models is fair use because the process is transformative: the models learn statistical patterns from text rather than storing books to republish them, and they serve a different purpose than the original works. In its motion against the news publishers, OpenAI said its expert found only a small number of verbatim reproductions of publishers’ content across a sample of 20 million ChatGPT conversation logs. Microsoft filed its own motion seeking judgment on the claims against it.

The companies received significant outside support on September 1, 2026, when the Justice Department filed a statement of interest in the MDL. The government argued that training models on written works is highly transformative and that requiring licenses for AI training generally would be legally wrong, and it cited national competitiveness and security interests. The statement addresses training, not model outputs. The news plaintiffs responded that the court should give it no weight because it was filed before the factual record was public and conflicts with the U.S. Copyright Office’s May 2025 report on generative AI training. A statement of interest does not bind the court.

Fair Use: The Legal Test That Will Decide the Case

Fair use, codified at 17 U.S.C. § 107, allows some unlicensed uses of copyrighted works. Courts weigh four statutory factors together rather than tallying them:

FactorWhat the court asksHow it plays out in AI training cases
1. Purpose and character of the useIs the use transformative, with a different purpose, and is it commercial?AI companies say training is a new, analytical use. Authors say the purpose is to produce competing writing for profit. The Supreme Court’s 2023 Warhol decision narrowed how far “transformative” can carry a commercial use.
2. Nature of the workIs the original creative or factual, published or unpublished?Novels are highly creative, which favors authors, though courts often give this factor little weight.
3. Amount usedHow much was taken, and was it reasonable for the purpose?Entire books were copied. Defendants argue full copies were necessary for training, as in the Google Books case.
4. Effect on the marketDoes the use harm the market for the original or for licenses?Usually decisive. Authors point to lost licensing revenue and to a flood of AI-generated books that dilutes demand for human work.

Because the OpenAI case sits in the Second Circuit, the court will also have to account for Authors Guild v. Google (2015), in which the Second Circuit held that Google’s scanning of millions of books to build a searchable index was fair use. OpenAI is expected to lean on that precedent. Authors argue the comparison fails because Google Books showed only snippets and did not generate substitute works.

How Other Courts Have Ruled on AI Training

No appellate court has ruled on whether training a generative AI model on books is fair use. The trial-level decisions so far point in different directions, and each turned on its own record.

Bartz v. Anthropic (N.D. Cal.)

In June 2025, U.S. District Judge William Alsup split the issue. He held that Anthropic’s use of lawfully purchased books to train its Claude models was fair use, but that building a permanent library of pirated copies downloaded from LibGen and similar sites was not. After he certified a class of rightsholders, Anthropic agreed to pay $1.5 billion, roughly $3,000 per work, without admitting liability. Judge Araceli Martínez-Olguín granted final approval on July 20, 2026, the largest copyright class settlement on record.

Kadrey v. Meta (N.D. Cal.)

Days later, U.S. District Judge Vince Chhabria ruled for Meta on fair use against 13 authors who claimed their books were used to train the Llama models. His reasoning rested on the authors’ failure to prove market harm on that record, and he cautioned that many uses of copyrighted works to train AI could be unlawful where the harm is shown. In July 2026 he declined the authors’ request for an immediate appeal, preferring to wait until the remaining issues could be reviewed together.

Thomson Reuters v. Ross Intelligence (3d Cir.)

On September 29, 2026, the U.S. Court of Appeals for the Third Circuit issued the first federal appellate decision on fair use in AI training. It affirmed that Ross, a now-defunct legal research startup, infringed Thomson Reuters’ Westlaw headnotes by using them to train a competing search tool, finding the use minimally transformative and harmful to the market. The court stressed that Ross was not a generative AI system and distinguished the OpenAI, Anthropic and Meta cases, so its reach is limited, but it is binding precedent in the Third Circuit and persuasive elsewhere.

CaseCourt / dateResultWhy it matters for OpenAI
Bartz v. AnthropicN.D. Cal., June 2025; settled 2026Training on purchased books fair use; pirated library notSeparates acquisition from training, the same line authors draw against OpenAI
Kadrey v. MetaN.D. Cal., June 2025Fair use on that record; market harm not provenShows how much the fourth factor depends on evidence
Thomson Reuters v. Ross3d Cir., Sept. 29, 2026Not fair useFirst appellate ruling; involves non-generative AI
In re OpenAI (authors)S.D.N.Y., pendingNo rulingFirst major generative AI fair use decision in the Second Circuit

Why the Piracy Allegations Matter

The Anthropic ruling changed the strategy in author cases. Even if a court agrees that training is transformative, acquiring books from pirate sites is a separate act of copying that may not be excused. That is why the authors’ brief against OpenAI focuses heavily on LibGen and on how the company handled the Books1 and Books2 datasets.

The stakes are financial. Under 17 U.S.C. § 504(c), a copyright owner may elect statutory damages of $750 to $30,000 per infringed work, and up to $150,000 per work if infringement is found to be willful, without proving actual losses. Across hundreds of thousands of books, even the low end produces very large numbers, which is the pressure that preceded Anthropic’s settlement. Statutory damages generally require that the work was registered with the U.S. Copyright Office before the infringement began or within three months of publication, under 17 U.S.C. § 412, so registration status can determine who is eligible for that remedy.

The evidence fight has been intense. In February 2026, Judge Stein sustained OpenAI’s objection to a magistrate judge’s order that would have required it to produce attorney communications about deleting the datasets, holding that denying willful infringement did not waive attorney-client privilege. Willfulness remains contested.

Class Certification: What It Means for Authors

The authors’ case is filed as a proposed class action on behalf of writers whose copyrighted books were allegedly used to train OpenAI’s models. A class has not been certified. According to the Authors Guild, the court agreed to decide whether a class can be certified only after it rules on the merits, so the case is moving forward on behalf of the named plaintiffs for now.

That sequencing matters. If OpenAI wins on fair use, class certification may never be reached. If the authors win on some claims, the class question, including how a class would be defined and whether registered works can be identified at scale, becomes the next major battle. In Anthropic, certification of a class covering pirated books came shortly before the settlement.

The News Publisher Track

The New York Times is the most prominent plaintiff in the parallel news track, joined by the Daily News papers, the Center for Investigative Reporting, The Intercept and Ziff Davis. Their cases are on the same summary judgment schedule. Publishers argue that OpenAI and Microsoft copied their journalism at multiple stages, from acquisition and training to retrieving articles to ground chatbot answers, and that the resulting products divert readers. In August 2026, after the Supreme Court narrowed contributory copyright liability in Cox Communications v. Sony Music Entertainment, Judge Stein dismissed the publishers’ “material contribution” claims and a trademark dilution claim, leaving their core direct infringement claims intact. A publishers’ motion for sanctions over preservation of ChatGPT logs is also pending.

New suits keep coming. USA Today Co. and 19 affiliated papers sued OpenAI and Microsoft on October 8, 2026, seeking more than $250 million, and that case has been paused by agreement until the pending summary judgment motions are decided.

What Happens Next

  1. Reply briefs, November 20, 2026. Each side responds to the other’s opposition. Motions to exclude expert testimony are also pending.
  2. Hearing. The parties have asked for oral argument. The Authors Guild expects a hearing in early 2027; none has been scheduled.
  3. Ruling. Judge Stein may grant judgment on some claims, deny it on others, or find disputed facts that require a jury. Different conclusions are possible for acquisition, training and outputs.
  4. Class certification and trial. Only claims that survive summary judgment move forward, followed by class certification briefing for the authors.
  5. Appeal. Any final ruling on fair use is likely to reach the Second Circuit, which could create a split with the Ninth Circuit district rulings and the Third Circuit’s Ross decision.

Settlement remains possible at any stage. Anthropic settled after a ruling on piracy exposure and a class certification order, and commentators have noted that OpenAI’s plans for a public offering could increase pressure to resolve the cases. OpenAI has not announced any settlement.

What Authors Should Know Now

The following is general information, not legal advice. Authors with questions about their own works or rights should consult a licensed copyright attorney.

  • There is nothing to claim yet. No class has been certified and no settlement exists in the OpenAI case. Be cautious about any message asking for payment or personal information to “join” it.
  • Check your registrations. Whether a book is registered with the U.S. Copyright Office, and when, can affect eligibility for statutory damages in any copyright case. Registration records are searchable through the Copyright Office.
  • Keep your publishing records organized. Contracts, rights reversions and ISBNs can determine who owns the claim to a given work and who would be paid in a future class process.
  • Review AI clauses in new contracts. Publishers and author groups have introduced contract language addressing AI training and licensing.
  • Follow official notices. If a class is certified or a settlement is reached, the court will approve a formal notice describing who is covered and the deadlines.

Readers with information about AI training practices or the publishing industry can reach our newsroom through the Legal News Report tip line. For related coverage, see our Business & Bankruptcy, Settlements, and Consumer Rights sections.

Frequently Asked Questions

What is the OpenAI copyright lawsuit about?

Authors, including the Authors Guild and novelists such as John Grisham, George R.R. Martin and Jodi Picoult, allege that OpenAI and Microsoft copied their copyrighted books without permission to train the GPT models behind ChatGPT, that some copies came from pirate websites, and that ChatGPT can produce outputs similar to their works. OpenAI and Microsoft deny infringement and argue their use is fair use. No court has decided the merits.

Where is the case and who is the judge?

The author class case and the news publishers’ cases are consolidated in multidistrict litigation, In re OpenAI, Inc. Copyright Infringement Litigation, MDL No. 3143 (No. 1:25-md-03143), in the U.S. District Court for the Southern District of New York before U.S. District Judge Sidney H. Stein.

Has a court ruled that AI training is fair use?

Not in this case. Judge Stein has not ruled on the summary judgment motions. Elsewhere, two federal judges in California found AI training on books to be fair use on the records before them in 2025, while the Third Circuit in September 2026 held that a legal research startup’s use of Westlaw headnotes to train a non-generative AI tool was not fair use. None of those rulings binds the court in New York.

Is there an OpenAI settlement authors can claim from?

No. As of October 28, 2026, there is no settlement and no claims process in the OpenAI case. The $1.5 billion settlement that authors may have heard about resolved a different case, Bartz v. Anthropic, and its claim deadline has passed.

Am I part of the authors’ class action against OpenAI?

No class has been certified. The court agreed to address class certification after deciding the merits, so for now the case proceeds on behalf of the named plaintiffs. If a class is later certified, notice would explain who is included and what options class members have.

When will the court decide?

Summary judgment reply briefs are due November 20, 2026 under the schedule reported from the court docket. The Authors Guild has said it expects a hearing in early 2027. Judge Stein could decide some or all of the claims on the motions or send disputed issues to a jury trial, and no trial date has been set.

Sources

Legal disclaimer: This article is for general informational and news purposes only and does not constitute legal advice or create an attorney-client relationship. Laws, court schedules and deadlines change. Allegations described in pending lawsuits are unproven claims, and OpenAI and Microsoft deny infringement. See our full legal disclaimer, editorial standards, and corrections policy.