In the Southern District of New York, the Authors Guild and a proposed class of novelists allege that OpenAI and Microsoft copied their books to train GPT models. Cross-motions for summary judgment are pending in MDL No. 3143, and largely unredacted briefs became public on September 17, 2026. Opposition briefs are due in early October 2026 and reply briefs in early November 2026.
Latest development Sep 17, 2026
On September 17, 2026, largely unredacted versions of the September 5 summary judgment briefs and supporting Rule 56.1 statements became public. The Authors Guild highlighted internal OpenAI communications that it says show executives discussing purging LibGen-sourced pirated books from training data. Opposition briefs are due in early October 2026.
Why it matters
The court will decide whether copying books to train GPT models is fair use, a question that sets OpenAI’s and Microsoft’s exposure to authors. Anthropic’s $1.5B settlement in Bartz v. Anthropic offers a benchmark, and output-based claims over ChatGPT’s plot summaries also remain live.
Background
In September 2023, the Authors Guild and 17 fiction writers filed a proposed class action in the U.S. District Court for the Southern District of New York. The writers include John Grisham, George R.R. Martin, and Jodi Picoult. They allege that OpenAI, with Microsoft’s backing, copied their books to train GPT models and that ChatGPT can generate infringing outputs derived from those works. Microsoft was later added as a defendant. In April 2025, the Judicial Panel on Multidistrict Litigation consolidated the case with other author and news actions into MDL No. 3143 before Judge Sidney H. Stein.
On October 27, 2025, the court denied OpenAI’s motion to dismiss the class complaint’s output-based infringement claim. It held that a “more discerning observer” could find ChatGPT’s detailed plot summaries substantially similar to the originals. The court worked through a ChatGPT summary of Martin’s A Game of Thrones and expressly reserved the fair use question. On December 15, 2025, the court dismissed some claims brought by Ziff Davis in the MDL but let its contributory copyright infringement and copyright management information claims proceed.
Class discovery followed. On January 5, 2026, the court affirmed OpenAI’s court-ordered production of 20 million de-identified ChatGPT logs. Those logs bear directly on how often outputs reproduce class works. On February 6, 2026, the court sustained OpenAI’s objection to a magistrate judge’s order. That order had found a waiver of attorney-client privilege connected to OpenAI’s deletion of the Books1 and Books2 training datasets.
Summary judgment briefing opened on September 4, 2026, when Microsoft moved against the class claims, arguing that Copilot rarely reproduces books. On September 5, 2026, the authors cross-moved, arguing that fair use does not protect OpenAI’s copying. Largely unredacted briefs became public on September 17, 2026, and the Authors Guild highlighted internal OpenAI communications that it says show executives discussing purging LibGen-sourced books. Opposition briefs are due in early October 2026 and reply briefs in early November 2026. Bellwether selection is expected in late 2026, against the backdrop of Anthropic’s $1.5B settlement in Bartz v. Anthropic.
Timeline 9 events
-
FilingLargely unredacted summary judgment briefs become public
Largely unredacted versions of the September 5 summary judgment briefs and supporting Rule 56.1 statements became public. The Authors Guild highlighted internal OpenAI communications it says show executives discussing purging LibGen-sourced pirated books from training data.
Unsealed Briefs in Authors’ Case v. Microsoft/OpenAI: Top Execs Knew Their Mass Book Piracy Was IllegalMostly Unredacted briefs in New York Times, Authors Guild suits v. OpenAI, Microsoft now availableUnsealed Files Show Open AI/Microsoft Knew Copying Was Illegal and Could Hurt Authors
-
FilingAuthors Guild and co-plaintiffs cross-move for summary judgment
The Authors Guild and 13 co-plaintiffs, including Jonathan Franzen and George R.R. Martin, moved against OpenAI and Microsoft. They argued that fair use does not justify OpenAI’s copying, which they said threatens the market for human-authored books.
Authors Guild, Co-Plaintiffs Seek Summary Judgment in OpenAI CaseAuthors Guild, Co-Plaintiffs Seek Summary Judgment in OpenAI Case
-
FilingMicrosoft moves for summary judgment against the class claims
Microsoft argued that Copilot rarely reproduces books, and OpenAI moved on the news plaintiffs’ claims. The parties were to re-file briefs publicly, with unchallenged portions unredacted, by September 17.
-
OrderJudge Stein finds no privilege waiver over deleted training datasets
Judge Stein sustained OpenAI’s objection to a magistrate judge’s order. That order had found a waiver of attorney-client privilege connected to OpenAI’s deletion of the Books1 and Books2 training datasets.
Judge Stein reverses Magistrate Judge's ruling on OpenAI's attorney-client privilegeIP Hot Topic: Privilege Preserved
-
OrderJudge Stein affirms production of 20 million ChatGPT logs
The ruling upheld discovery central to proving how often ChatGPT outputs reproduce class works.
-
OrderCourt partly dismisses Ziff Davis claims within the MDL
The court granted OpenAI’s motion to dismiss some claims. It denied the motion as to Ziff Davis’s contributory copyright infringement and copyright management information removal claims.
Ziff Davis v OpenAI: Key Copyright Litigation RulingIn re OpenAI, Inc. Copyright Infringement Litigation
-
OrderCourt holds ChatGPT plot summaries could infringe
Judge Stein denied OpenAI’s motion to dismiss the authors’ output claims, finding a discerning observer could deem detailed plot summaries substantially similar. The court expressly reserved fair use.
-
OrderJudicial panel consolidates case into MDL No. 3143
The Judicial Panel on Multidistrict Litigation centralized the author and news-publisher suits before Judge Stein in the U.S. District Court for the Southern District of New York.
-
FilingAuthors Guild and 17 novelists sue OpenAI
The proposed class action alleged that OpenAI copied books to train GPT models. Microsoft was later added as a defendant.
Experimental AI forecast
How it is madeHow will the court resolve the cross-motions for summary judgment on fair use in the author class claims against OpenAI and Microsoft by September 30, 2027?
- Split ruling or fair use left open35%
- Still pending on September 30, 202725%
- Training held fair use15%
- OpenAI settles with authors15%
- Authors defeat fair use defense10%
- Resolves by
- Sep 30, 2027in 11 months
- Record confidence
- Medium
A split ruling, or a denial that leaves fair use for trial, is the most likely result at 35% by September 30, 2027. The authors’ evidence on LibGen-sourced books mirrors the pirated-library issue that produced mixed results in earlier AI training cases. Briefing ends in November 2026, leaving time for a ruling.
How it resolves. Resolves on the first of these events to occur on or before September 30, 2027, per the S.D.N.Y. docket or reliable reporting. “Training held fair use”: the court grants OpenAI and Microsoft summary judgment that all training-related copying in the author class claims is fair use. “Authors defeat fair use”: the court grants the authors summary judgment rejecting the fair use defense for training-related copying. “Split ruling or fair use left open”: the court rules on the motions but finds fair use for some copying and not others, or denies both sides’ motions on fair use, leaving it for trial. “OpenAI settles with authors”: before any such ruling, the class plaintiffs and OpenAI announce a signed or court-filed settlement of the class claims, whether or not it is approved. “Still pending”: none of the above has occurred by September 30, 2027.
Reasoning
Starting point. Cross-motions for summary judgment on fair use in U.S. copyright cases over AI training, including the 2025 rulings in Bartz v. Anthropic and Kadrey v. Meta: Judgment, not a measured frequency: cross-motions on fair use rarely yield a clean win for one side. Both 2025 AI training rulings produced partial or narrow results. Complex MDL summary judgment rulings often arrive six to twelve months after full briefing.
Cross-motions on fair use seldom end with a clean win for either side. In the 2025 Bartz v. Anthropic ruling, training was held fair use but pirated library copies were not. Kadrey v. Meta also turned on narrow, record-specific findings about market harm. Here, output claims survived dismissal on October 27, 2025, and the authors now press LibGen evidence resembling the Bartz piracy issue. That pattern makes a split ruling or a denial for fact disputes the likeliest result, at 35%. Reply briefs are due in early November 2026, but the court must also handle the news plaintiffs’ motions, so 25% remains on still pending. Settlement stays at 15% given the $1.5 billion Anthropic benchmark and bellwether pressure, offset by the defendants’ resources and precedent concerns. No development since the September 25, 2026, forecast moved any outcome by five points, so the numbers are unchanged.
Pirated LibGen sourcing evidence invites a split on fair use · Output claims already survived dismissal in October 2025 · Briefing closes November 2026, ruling timing uncertain · Anthropic’s $1.5 billion settlement sets a settlement benchmark · Defendants’ resources and precedent stakes favor litigating
Generated automatically by Claude Opus 5.5 on Sep 27, 2026 from this case’s record on this site, and published without editorial review. The probabilities are the model’s judgment, not measured frequencies, and they are scored publicly as cases resolve. Not legal or investment advice. How the forecasts are made
Disclosure: Anthropic, which makes the model that wrote this forecast, competes with or has commercial ties to a party to this case.