The New York Times lawsuit against OpenAI and Microsoft, filed in December 2023, reached its most consequential procedural phase in September 2026. All parties filed motions for summary judgment on whether training ChatGPT on millions of news articles qualifies as fair use under U.S. copyright law. The outcome could determine whether news publishers can block large-scale training or must pursue licensing instead. Parallel author class actions in the same multidistrict litigation face the same fair use question with different market-harm theories.
This tracker covers NYT OpenAI copyright appeal developments in 2026, summarizes core OpenAI training data lawsuit arguments, and explains generative AI fair use stakes for publishers and buyers of AI writing tools. Compare vendor posture with AI copyright resources as licensing markets evolve.
NYT vs OpenAI Case Background and Stakes
The Times alleges OpenAI and Microsoft copied millions of copyrighted news articles to train ChatGPT, producing outputs that compete with Times journalism and dilute its brand. The complaint joins separate news publisher suits consolidated before U.S. District Judge Sidney Stein in the Southern District of New York. Author plaintiffs including John Grisham and George R.R. Martin pursue parallel claims in related MDL dockets.
Unlike California training cases decided in 2025 and 2026 (Bartz v. Anthropic, Kadrey v. Meta), the Times emphasizes market substitution: ChatGPT answers current-events questions using Times-style reporting without licensing fees. The Times argues Second Circuit and Supreme Court precedents on transformative use should govern, not recent California district court rulings that found training fair use when plaintiffs did not prove output competition.
OpenAI and Microsoft counter that training is highly transformative, that facts and ideas in news are not protectable, and that licensing every training work is impractical. They cite Kadrey and Bartz while distinguishing output infringement claims that survived earlier motions to dismiss, including allegations that ChatGPT reproduced protectable expression from works like Game of Thrones.
2026 Procedural Updates and Summary Judgment Briefing
Parties filed opening summary judgment briefs on September 4, 2026, with oppositions due October 9, 2026, and replies due November 6, 2026. Judge Stein set word limits separately for news cases and consolidated class cases. Daubert expert motions follow a parallel schedule with openings due September 24, 2026.
August 2026 brought significant procedural rulings. The Supreme Court's Cox Communications decision held that "material contribution" is not a valid theory of contributory copyright infringement. Judge Stein dismissed with prejudice news plaintiffs' contributory claims premised on material contribution and end-user direct infringement. The court also dismissed trademark dilution claims by the Times and Daily News and denied motions to amend complaints with new contributory theories against Microsoft.
Output infringement claims remain live. Judge Stein previously denied OpenAI's motion to dismiss output claims where plaintiffs alleged ChatGPT reproduced protectable summaries. The September 2026 fair use briefing therefore targets training liability first, but output theories could still proceed to trial if summary judgment is denied on those claims.
Core Legal Arguments on Fair Use and Market Harm
The fair use fight centers on whether LLM training is a transformative purpose that outweighs market harm to publishers. OpenAI argues training creates a new functional technology that does not supersede news articles as reading experiences. The Times argues training built a commercial substitute that answers questions using Times-derived knowledge without compensation.
The U.S. Department of Justice filed a statement of interest in September 2026 supporting OpenAI. DOJ called AI training an "extraordinarily transformative" fair use purpose and argued a novel "dilution" market-harm theory advanced by some plaintiffs is deeply flawed. DOJ also cited national security and competitiveness interests in U.S. frontier model development.
| Party | Fair use position | Market harm theory |
|---|---|---|
| OpenAI / Microsoft | Training is transformative; facts not copyrightable | No proven substitution; licensing impractical at scale |
| NYT / news plaintiffs | Training copies expressive works to build competing product | ChatGPT substitutes for Times reporting on current events |
| U.S. government (SOI) | Extraordinarily transformative fair use for LLM development | Reject dilution theory; balance public benefit |
Settlement and Licensing Signals for Publishers
While the Times case proceeds, the broader market moved toward licensing even as courts debated fair use. OpenAI, Google, and other labs announced content partnerships with news and media companies throughout 2025 and 2026. Licensing deals do not resolve the legal question for unlicensed historical training data, but they signal commercial settlement paths if judges split outcomes between training and outputs.
Anthropic's $1.5 billion Bartz settlement over pirated shadow-library copies shows that acquisition method matters independently of fair use. Publishers weighing litigation versus deals should model both training fair use risk and storage or scraping liability. Enterprise buyers of AI writing tools should require vendor indemnities, training data disclosures, and output ownership terms that reflect unsettled law.
Judge Stein may rule on summary judgment in late 2026 or early 2027. A trial on surviving claims could follow in 2027 if motions are denied. Second Circuit appeal would likely follow any major fair use ruling, prolonging uncertainty for publishers and model vendors.
FAQ for Publishers and AI Buyers
Is AI training fair use decided in the NYT case?
Not yet. Summary judgment briefing began September 2026. California rulings in Kadrey and Bartz favor training fair use on different records, but the Times case is in the Second Circuit's backyard with distinct market-harm allegations.
Why did DOJ intervene?
DOJ filed a statement of interest arguing training LLMs is extraordinarily transformative fair use and citing national security interests in U.S. AI leadership. Statements of interest are not binding rulings but signal executive branch views.
What happened to contributory infringement claims?
Judge Stein dismissed material-contribution contributory claims after Cox Communications. Amended contributory theories against Microsoft were denied in August 2026.
Do output infringement claims remain?
Yes. Judge Stein allowed output claims to proceed where plaintiffs alleged reproduction of protectable expression, separate from training fair use.
What should enterprise AI buyers do now?
Document vendor training policies, negotiate indemnities, monitor case outcomes, and prefer vendors with documented licensing for news and book corpora where substitution risk is high.
