On Friday, the Wikimedia Foundation, the nonprofit behind Wikipedia, filed a neutral friend-of-the-court brief in the OpenAI copyright lawsuit in Manhattan. The filing asks the court to separate two questions the case has tangled together: whether training an AI model on copyrighted works is fair use, and whether the model can reproduce those works in its answers. Wikimedia is not a party to the case and is not backing either side.
The timing is deliberate. Summary-judgment briefing on fair use is still underway before Judge Sidney Stein in the Southern District of New York, and he has not ruled. Wikimedia, which says it has a large stake in the open web, wants its view in the record before he does.
What Wikimedia asked the judge to do
Wikimedia says both sides push fair use too far or describe it too simply. It is not siding with OpenAI or with The New York Times. Its position, laid out in the foundation's blog post and in press coverage of the filing, comes down to three points:
- Training a model on copyrighted works without a license may be fair use when the resulting model is something new.
- Scraping websites should not, on its own, count as infringement.
- When a model returns copies that substitute for readers' engagement with the original works, a license should be required.
The blog post describes the goal as output that is “something new and different from the original sources.” That test shifts the argument away from whether a system read a page and toward whether the system hands users a stand-in for it. Mezha's report on the filing describes the same framing.
I could not read the brief itself, so this account relies on the blog post and press coverage. The substitution distinction is the piece to watch. A judge could adopt it without deciding the case as a whole, which makes it the most likely part of the filing to matter in practice.
Why the fair-use ruling is still pending
The New York Times sued OpenAI and Microsoft in December 2023, alleging that the companies used its articles to build products that compete with the paper. Wikipedia's entry on the case tracks the history. Judge Stein rejected a bid to dismiss in March 2025, but he has not decided the fair-use question. The consolidated case now bundles 12 lawsuits, according to Wikimedia's account of the filing.
The key dates so far:
| Date | What happened |
|---|---|
| December 2023 | The New York Times sues OpenAI and Microsoft |
| March 2025 | Judge Stein rejects a bid to dismiss |
| September 4, 2026 | Both sides file cross-motions for summary judgment on fair use |
| September 29, 2026 | Third Circuit affirms a ruling rejecting fair use for directly competing AI training |
| October 23, 2026 | Oppositions filed |
| October 30, 2026 | Wikimedia files its amicus brief |
| November 6 or 20, 2026 | Reply deadline, reported differently by sources |
The reply deadline is where the sources disagree. The AI Lawsuit Tracker and Axis Intelligence's tracker put it on November 6, 2026. OpenClassActions and Legal News Report say November 20. The court docket is the authority here, and it should be checked before anyone plans around either date.
The Times's allegations have not been proven. OpenAI and Microsoft deny infringement, according to OpenClassActions.
Where the Justice Department and conservative groups stand
The Justice Department has filed a statement of interest on OpenAI's side, arguing that training large language models falls within fair use. CryptoBriefing reports that filing as dated September 2, while AI Copyright Map dates it September 1. The sources have not reconciled the two dates.
The Center for American Rights, a conservative nonprofit law firm, filed its own amicus brief this week. It warns against an expansive reading of fair use, and the brief includes this line: “If the Court here agrees with OpenAI’s expansive view of fair use,” according to coverage of the filing. The group argues that broad crawling would harm news outlets and national security.
OpenAI's public position is that training is transformative and does not substitute for the original works. The company lays out that framing on its page about the Times lawsuit.
Wikimedia sits between these camps. It criticizes the pro-OpenAI and pro-publisher positions alike as too broad or too simple, and it proposes a middle path that turns on what a model produces. That makes the question narrower than a blanket verdict on training, and a judge may find a narrower question easier to answer.
What the Third Circuit ruling changes
On September 29, 2026, the Third Circuit affirmed a ruling that rejected fair use for AI training that directly competes with the source product, according to Axis Intelligence and the AI copyright lawsuit tracker. It is the appellate guidance the trackers point to most often on this question.
The ruling does not bind a federal judge in Manhattan, who sits in the Second Circuit. Judges still read appellate reasoning closely, and both sides are likely to cite it. The Third Circuit's focus on direct competition gives Wikimedia's substitution test a concrete reference point, which is one reason the nonprofit's proposal may get more attention now than it would have a year ago.
The wider docket and the separate newspaper suit
The docket is crowded beyond this case. Trackers count 144 AI copyright cases nationally, with 113 still active or pending, according to the AI copyright lawsuit tracker.
On October 8, 2026, 19 affiliated papers sued OpenAI and Microsoft and sought more than $250 million, according to Legal News Report. That case is paused until the judge resolves the summary-judgment motions. The pause ties the second suit to the same ruling that Wikimedia is trying to shape.
Microsoft is a defendant alongside OpenAI, and its financial ties to the company are large. USA 24 has reported that Microsoft booked $10 billion from OpenAI in a single quarter, about 11% of its revenue.
What readers and creators should do now
Most readers need to do nothing today. Three groups can act now:
- Publishers and creators should check the consolidated docket directly for the reply deadline, rather than relying on third-party trackers that disagree. They should also review their licensing terms and crawling rules, including robots.txt directives, so they know what they have allowed.
- Businesses that build on chatbot output should treat the legal status of generated text as unsettled. If an output reproduces an article, verify it and keep the source.
- Consumers of ChatGPT or Microsoft Copilot need no action now. Product or content changes could follow a ruling, but no timetable has been set.
What comes next
Judge Stein's decision is the next event that matters. Watch whether he separates training from outputs, as Wikimedia urges, or rules on fair use more broadly. Also watch whether he treats Microsoft's Copilot outputs differently from OpenAI's training, a question the case leaves open, according to AI Copyright Map.
Congress is not the only venue for AI rules. AI regulation has become a midterm election issue, and a fair-use ruling could move part of that debate into federal courtrooms. Whatever Stein decides, the first answer will come from a judge in Manhattan, not from a legislature.
