Skip to main content

Case brief · S.D.N.Y. · fair use undecided as of Oct 29, 2026

NYT v. OpenAI: Where the Case Stands on October 29, 2026

No court has ruled on fair use in the New York Times case against OpenAI and Microsoft. The court has decided which claims go forward, and the parties are now arguing fair use on a full record. This page dates every step and says which document or outlet each one comes from.

Direct answer

As of November 20, 2026, in the MDL docket entries we read (through ECF 2415, filed October 30, 2026), no court has ruled on fair use in NYT v. OpenAI. Public redacted versions of the summary judgment oppositions were filed on October 29, 2026 (ECF 2309 to 2347). Judge Sidney Stein let the Times’ direct and contributory infringement claims proceed on April 4, 2025. The parties filed summary judgment papers in September 2026, and the Justice Department has backed OpenAI. The decision is pending in the entries we read.

AI Vortex case map showing the New York Times v OpenAI litigation timeline and the DOJ filing's separation of acquisition and storage, training, and output questions

AI Vortex / Decision map

The DOJ filing changes the argument, not the ruling

  1. Acquisition / storage
  2. Training
  3. Outputs
  4. DOJ position
  5. Court posture
  6. Decision pending
Decision map. The filing advocates a fair-use position for training-stage copying while leaving acquisition, storage and output questions distinct and the court's ruling pending.

As of October 31, 2026, the court’s own April 4, 2025 opinion in the New York Times case against Microsoft and OpenAI, re-read today, decided only motions to dismiss and never mentions fair use. No fair use ruling appears in the sources we read. We could not re-read the docket today, so later steps rest on the sources dated below.

Who this page is for

This page is for publishers, AI companies, counsel and policy teams who need the status of the case from dated documents, not a prediction.

Decision framework

  • Use this page if: you need to know which claims survived, what the parties now dispute and which document backs each step.
  • Do not use it as: a forecast of the fair use ruling or a substitute for the live docket before you act on timing.
  • Next step: read the timeline below, then check the MDL docket for anything filed after October 30, 2026.

Source and date: the April 4, 2025 opinion on the Southern District of New York’s own site, read on October 31, 2026 (it denies the motions to dismiss the contributory infringement claims and the direct infringement claims over older conduct, and contains no fair use ruling). The MDL docket PDFs were read in the public RECAP archive of the docket on November 20, 2026, through ECF 2415 (October 30, 2026). Nothing filed after October 30, 2026 is described here; check the live docket for later entries. Related: AI copyright training-data landscape.

Current status of NYT v. OpenAI, as of October 29, 2026

The New York Times Company sued Microsoft and OpenAI in the Southern District of New York on December 27, 2023, and that case now sits inside a consolidated proceeding, In re OpenAI, Inc. Copyright Infringement Litigation, 25-md-3143, before Judge Sidney H. Stein. The fair use question, which decides whether training a large language model on news articles needs a license, has not been answered by the court. The court has only decided which claims may go forward.

What we can document from the sources we read is this. Summary judgment is the current stage, and the defendants' oppositions are now in: on October 23, 2026 OpenAI and Microsoft filed their oppositions to the news plaintiffs' and the book plaintiffs' summary judgment motions, under seal, and asked Judge Stein for oral argument (MDL docket, entries 2232, 2234 and 2236). The docket entries we read show no hearing date and no ruling. Microsoft’s public redacted memorandum on the news plaintiffs’ cases is Doc. 1975, filed September 17, 2026, and a separate class plaintiffs’ memorandum for partial summary judgment, from the authors’ side of the MDL, is Doc. 1982, filed the same day. A September 15, 2026 report syndicated on Yahoo Finance, which cites Axios, says each side has laid out its arguments to Judge Stein and that he is expected to decide whether all or part of the case goes to trial in 2027. We did not find a fair use ruling in any source we read, so the Decision pending label on this page is our reading of those sources, not a docket entry.

This page owns the NYT v. OpenAI docket, timeline and open questions. The cross-case AI copyright training-data landscape covers how other courts have treated training.

Timeline of the case, with the source for each step

DateWhat happenedSource
Dec. 27, 2023The Times files suit against Microsoft and OpenAI asserting direct, vicarious and contributory copyright infringement, DMCA violations, unfair competition by misappropriation and trademark dilution.April 4, 2025 opinion
Apr. 30, 2024Eight Daily News-group newspapers file a parallel complaint with the same claims plus a New York dilution claim. The Center for Investigative Reporting files in June 2024.April 4, 2025 opinion
Aug. 2024The Times files a First Amended Complaint with more works but no new causes of action.April 4, 2025 opinion
Jan. 14, 2025Oral argument on the motions to dismiss, cited in the opinion as the transcript at ECF No. 433.April 4, 2025 opinion
Apr. 3, 2025The JPML centralizes twelve actions from the Northern District of California and the Southern District of New York in the Southern District of New York.JPML Transfer Order
Apr. 4, 2025Judge Stein rules on the motions to dismiss: contributory infringement, the 2019 and 2020 conduct and trademark dilution survive; other claims are narrowed.April 4, 2025 opinion
Spring 2025Case Management Order No. 1 sets the initial conference for May 22, 2025 and stays deadlines until then.CMO No. 1
Jan. 5, 2026Judge Stein affirms Magistrate Judge Wang’s orders of Nov. 7 and Dec. 5, 2025 that OpenAI produce its entire 20 million de-identified ChatGPT log sample to the news and class plaintiffs, and denies OpenAI’s objections.Order of Jan. 5, 2026
June 25, 2026After the Supreme Court’s Cox v. Sony decision, the Times asks leave to amend: it would voluntarily dismiss its alternative contributory infringement claim against OpenAI and Microsoft (Count IV) and its trademark dilution claims (Count VII), and amend its contributory claim against Microsoft (Count III). The filing says OpenAI did not oppose the two dismissals.Doc. 1590-1
July 9, 2026The Times and other news plaintiffs move for sanctions against OpenAI over discovery conduct, according to Bloomberg Law. The motion itself is not among the documents reviewed here, and no ruling on it appears in the sources reviewed.Bloomberg Law
Aug. 6, 2026The court rejects the Times’ attempt to add a claim that Microsoft encouraged infringement, as too late, according to Bloomberg Law.Bloomberg Law
Sept. 4, 2026Both sides in the MDL were to file their motions for summary judgment that day, according to a litigation tracker. The memoranda we read carry the Sept. 17 filing date below.chatgptiseatingtheworld.com
Sept. 15, 2026A report citing Axios says both sides have presented their arguments and that the Justice Department filed a brief backing OpenAI. This is the DOJ position as reported.The Cool Down via Yahoo Finance
Sept. 17, 2026Microsoft files its public memorandum for summary judgment on the news cases (Doc. 1975), and the authors’ class plaintiffs file theirs (Doc. 1982).The two memoranda
Oct. 23, 2026OpenAI and Microsoft file, under seal, their oppositions to the news plaintiffs' and the book plaintiffs' summary judgment motions, and each files a letter motion asking for oral argument (entries 2232, 2234, 2236). The class plaintiffs file sealed opposition papers on Oct. 23 and 24 (entries 2260, 2262), with redacted versions on Oct. 24 (entries 2268, 2271).MDL docket on CourtListener
Oct. 28, 2026OpenAI files a letter motion to seal materials in its summary judgment opposition papers (entry 2300); responses to the sealing requests are filed the same day (entries 2284, 2287). On Oct. 29 publisher associations move for leave to file an amicus brief (entry 2304).MDL docket on CourtListener
Oct. 29, 2026Public redacted versions of the oppositions go on the docket. OpenAI files its opposition to the news plaintiffs’ motion for partial summary judgment (entry 2314) and to the class plaintiffs’ motion (entry 2312); Microsoft files its oppositions (entries 2309 to 2311). The brief in entry 2314 argues, in its headings, that OpenAI’s use was highly transformative and that the news plaintiffs have not suffered market harm. The news plaintiffs file an omnibus opposition to the defendants’ summary judgment motions (entry 2320), whose table of contents argues that copying to train LLMs and copying for retrieval-augmented generation (RAG) are not fair use. The class plaintiffs file public redacted oppositions to Microsoft and OpenAI (entries 2346, 2347), and plaintiffs oppose the defendants’ Rule 702 motion to exclude expert testimony (entry 2337).MDL docket PDFs (RECAP), read Nov. 20, 2026
Oct. 30, 2026Bloomberg Industry Group files a reply in support of its motion to intervene for the limited purpose of unsealing summary judgment sealing proceedings (entry 2400). Judge Stein dismisses a proposed amicus letter motion that did not request leave to file (entry 2412). Proposed orders on other amicus motions appear (for example entry 2415); we did not see a ruling on them.MDL docket PDFs (RECAP), read Nov. 20, 2026
Nov. 20, 2026Court posture: fair use undecided in the entries we read through October 30, 2026.Our reading

A note on what changed from the previous version of this page. The November 20, 2026 update adds the October 29 and 30 filings in the timeline above, read as PDFs in the public RECAP archive of the MDL docket. The October 29 public redacted briefs replace our earlier statement that the opposition papers were not public; what we say about them comes from their captions and tables of contents, so the arguments are described only at that level. Docket entries after ECF 2415 are not covered. The page described a Justice Department Statement of Interest by date and docket number; we did not re-read that filing, so the DOJ position is still described only as reported by the Sept. 15, 2026 article above.

What the parties dispute: fair use, training and outputs

The April 4, 2025 opinion lays out the structure everything else hangs on. According to the complaints, as the court summarized them, the defendants’ models touch the plaintiffs’ works at two stages. At the training stage the defendants collect and store content, build datasets from it and run it through the model to adjust parameters. At the output stage the models respond to prompts, and the plaintiffs say some responses reproduce their articles verbatim or nearly verbatim. The court also described a third route, where Microsoft’s Copilot and Browse with Bing combine GPT technology with search results and, the plaintiffs say, generate paraphrases and direct quotes without sending readers to their sites.

StageWhat the plaintiffs allegeWhat the defendants say
Acquisition / storageScraped content, including the Times’ articles, was collected into datasets such as WebText and Common Crawl and stored as copies.The opinion records OpenAI’s limitations defense for conduct in 2019 and 2020, which the court rejected at the pleading stage.
TrainingCopying articles to train GPT models infringes and is not excused by fair use.Copyright law permits training on public material and training is transformative; OpenAI points to Kadrey v. Meta and Bartz v. Anthropic, according to the Sept. 15 report.
OutputsModels regurgitate articles and substitute for the Times’ journalism, which harms subscriptions and advertising.Microsoft says 8.2 million chat logs show Copilot virtually never displays even a sentence of news content (Doc. 1975).

Training

The DOJ position, as reported, is that training large language models on millions of works creates a “transformative public benefit” that outweighs competitive harm to publishers. That is advocacy by a non-party in active litigation, not a holding. Microsoft’s brief makes the matching argument from the defense side: using works to build a model that captures human language is “new and different” from any earlier use, and that conclusion applies equally to news articles, it says, as it does to books.

Outputs and substitution

The Times’ answer, as the September report describes it, is that the California decisions involved different facts because those plaintiffs did not show market competition from outputs that substitute for their work. The Times says the systems act as a commercial substitute and that the court should follow Supreme Court and Second Circuit precedent. This is where the case differs most from the book cases. The JPML noted that the news plaintiffs assert infringement both when articles are used as input and in the outputs, which it described as verbatim and detailed summaries of news content.

Web grounding

Microsoft’s memorandum isolates a feature it calls web grounding, which pairs Copilot with Bing search. It argues the process uses intermediate copies that users do not see, that any site owner can opt out, and that discovery shows no market substitution. Those are Microsoft’s arguments. The news plaintiffs’ opposition to them is outside what we read.

What Judge Stein already decided on April 4, 2025

The opinion was not a fair use ruling. It decided which claims could survive a Rule 12(b)(6) motion, taking the complaints’ allegations as true. The result was mixed, which is why we avoid describing it as a win for either side.

  • Time bar: the court denied OpenAI’s motion to dismiss direct infringement claims based on conduct more than three years before the complaints were filed.
  • Contributory infringement: denied in all three actions. The court wrote that the plaintiffs had “plausibly alleged the existence of third-party end-user infringement and that defendants knew or had reason to know of that infringement.”
  • Trademark dilution: the motions to dismiss the federal and New York dilution claims in the Daily News action were denied.
  • Unfair competition by misappropriation: dismissed with prejudice as preempted by section 301 of the Copyright Act.
  • CIR “abridgment” claims: dismissed because the court found the summaries not substantially similar to the articles as a matter of law.
  • DMCA section 1202: dismissed without prejudice against Microsoft and, in the Times action, against OpenAI under section 1202(b)(1). The 1202(b)(3) claims were dismissed everywhere. OpenAI’s motions on the 1202(b)(1) claims in the Daily News and CIR actions were denied.

The opinion also records what the Times alleges about itself: more than 10 million registered works and nearly 10.1 million paid subscribers, and billions of dollars invested in journalism. These are allegations as summarized by the court, not findings.

What has happened since, and what is still moving

The JPML order explains why the case now reads as one proceeding. It found common questions of fact about how defendants trained their models and noted that defendants had already produced petabytes of training data. Judge Stein’s Case Management Order No. 1 then set the first conference for May 22, 2025 and stayed deadlines until it. On August 6, 2026, Bloomberg Law reported that the Times’ bid to amend its complaint against Microsoft with an inducement theory was denied as too late, and that the judge said a March 2026 Supreme Court decision on contributory liability did not change whether those theories were viable. We read only the visible opening of that article, so we describe it as reported and not as the order itself.

Two earlier records we did read in full explain parts of that fight. On January 5, 2026, Judge Stein affirmed the orders requiring OpenAI to hand over the whole 20 million-log sample, writing that Judge Wang’s rulings “were neither clearly erroneous nor contrary to law” (order). On June 25, 2026, the Times asked to drop its alternative contributory claim against OpenAI and Microsoft and its dilution claims and to recast the contributory claim against Microsoft after Cox (Doc. 1590-1). Bloomberg Law reported on July 9, 2026 that the Times and other news plaintiffs also moved for sanctions against OpenAI, asking that it be barred from relying on the output logs it produced (report). That motion is not among the documents reviewed here, no ruling on it appears in them, and none of them gives a reply-brief schedule.

The September 2026 memoranda show where the fight now sits. Microsoft asks for summary judgment on training and on web grounding. The authors’ class plaintiffs, whose books case runs on the same MDL docket, filed their own motion the same day. A decision on the news cases is the next event that would change this page.

Why it matters for law firms

Our reading, not a prediction: the case will not settle by itself whether training on copyrighted text is lawful in general. Courts in California have already applied fair use to training on thin or different records. The Times case tests the same doctrine against a plaintiff that claims output substitution and has produced discovery on it. A ruling here will say more about outputs and retrieval features than about training alone.

For counsel advising publishers, preserve separate evidence about how content was acquired, what was licensed, how it appears in outputs and what it does to subscriptions or traffic. For counsel advising AI developers, the filings show that dataset provenance, retention and the commercial effect of a use are litigated facts, and that a retrieval feature like web grounding gets its own analysis. For a firm that publishes online, the practical point is smaller: your own site is part of this training question, so know which crawlers your terms and robots.txt address.

Avoid two shortcuts. “The DOJ says training is legal” overstates a position paper. “The court let the case proceed” overstates a pleading ruling. The supportable statements are narrower: the government has taken a side on training as reported, and the court has not decided fair use.

What to record next

  1. Attach the April 4, 2025 opinion and the two September 17, 2026 memoranda to the matter file, and note which proposition each supports.
  2. Keep party allegations, judicial holdings, DOJ advocacy and press reports in separate fields.
  3. Map the exact dataset or content route to acquisition, storage, training and output records.
  4. Record market-harm and output questions as Unknown until the docket supports a conclusion.
  5. Set a review trigger for Judge Stein’s summary judgment decision and re-read the primary document the day it issues.

FAQ

What is the status of NYT v. OpenAI right now?

As of October 29, 2026, the case is at the summary judgment stage in the Southern District of New York before Judge Sidney H. Stein, inside the consolidated proceeding 25-md-3143. Microsoft’s public memorandum is dated September 17, 2026. None of the sources we read reports a ruling on fair use.

Has any court ruled that training on news articles is fair use in this case?

No ruling is reported in the sources we read. The April 4, 2025 opinion decided motions to dismiss under Rule 12(b)(6), not fair use. It let the contributory infringement claims, the older conduct and the dilution claims proceed and narrowed other claims.

What did Judge Stein decide on April 4, 2025?

He denied the motions to dismiss the contributory infringement claims and the time-barred argument for 2019 and 2020 conduct, denied the dilution motions in the Daily News action, and dismissed the unfair competition claims with prejudice as preempted. He also dismissed several DMCA claims without prejudice and the CIR abridgment claims.

When was the New York Times lawsuit filed, and against whom?

The Times filed its complaint on December 27, 2023, against Microsoft and OpenAI entities in the Southern District of New York. The April 4, 2025 opinion lists six claims, including direct, vicarious and contributory copyright infringement and DMCA violations.

Why is the case part of an MDL?

On April 3, 2025, the Judicial Panel on Multidistrict Litigation centralized twelve actions in the Southern District of New York because they share questions about how OpenAI and Microsoft trained their models. The news cases, the authors’ class actions and the DMCA cases now proceed before one judge.

What has the Justice Department said about the case?

A September 15, 2026 report citing Axios says the Justice Department filed a brief backing OpenAI and arguing that training creates a transformative public benefit. This is the government’s advocacy in active litigation, not a court holding, and we did not re-read the filing itself.

What does the case mean for lawyers using or advising on AI?

It shows that dataset provenance, licensing, retention and output behavior are litigated facts. Advisers to publishers and AI developers should document them separately and treat any ruling as a decision on that record, not a general rule for all AI training.

Related AI copyright and IP cases: RIAA labels v. Suno and Udio, Andersen v. Stability AI, Getty Images v. Stability AI (UK), Thaler v. Vidal (AI inventorship), Doe v. GitHub (Copilot).

Sources checked

Operational information, not legal advice. Verify current terms, account configuration and applicable professional duties before use.