Copyright Litigation Update: NYT v. OpenAI Survives Motion to Dismiss

copyright fair-use litigation June 24, 2025 · by Daniel Okonkwo

Judge Stein issued the long-awaited MTD ruling in The New York Times Co. v. OpenAI, Inc. on June 11. The headline: most of the Times's claims survive. Direct copyright infringement, contributory infringement, vicarious infringement, and the DMCA Section 1202 claims all proceed. The trademark dilution claim was dismissed. The case now moves to merits discovery on the fair-use defense and the proportionate-share questions that will dominate the second phase.

The ruling is 47 pages and worth reading in full for anyone advising clients on training-data exposure. Below, my read of the key passages and what they tell us about how the doctrine is going to land.

The headline holdings

1. Statute of limitations. Judge Stein adopts the discovery rule for copyright accrual, holding that the Times's claims accrued when it discovered (or reasonably should have discovered) the training-data use, not at the moment of training. This had been one of OpenAI's strongest defenses on the older works in the corpus, and its rejection meaningfully expands the damages exposure. Note: the Second Circuit's recent en banc treatment of Sohm v. Scholastic was the controlling guidance here, and the holding follows from that.
2. Direct infringement. The Times adequately alleged copying-in-fact through training and through the model's regeneration of substantial portions of Times articles in response to certain prompts. The "memorization" theory of regeneration-based infringement survives.
3. Contributory infringement. The court allows the Times's contributory claim to proceed against OpenAI on the theory that OpenAI knew or should have known that ChatGPT users were using the product to access Times content paywall-free. The framing of "should have known" carries weight; OpenAI's alleged failure to address known regeneration behaviors is enough to support the inference at the pleading stage.
4. DMCA Section 1202. The 1202(b) claims for removal of copyright management information survive. This is the sleeper issue in the broader litigation; statutory damages of $2,500-$25,000 per violation under Section 1203(c)(3) can produce eye-watering aggregate exposure independent of the underlying infringement claims. The court rejected OpenAI's argument that 1202(b) requires identical-bit copying that strips CMI, holding that functionally equivalent removal during training also qualifies.
5. Trademark dilution. Dismissed. The court held that the Times's trademark claims essentially restate the copyright claims and are preempted by Section 301 to the extent not, and otherwise fail on the merits because OpenAI's allegedly diluting use is not "in commerce" in the sense the Lanham Act requires.

What the ruling does not decide

The big one: fair use. Judge Stein expressly notes that fair use is a fact-intensive defense generally unsuitable for resolution at the pleading stage and reserves it for summary judgment. That is not a holding on the merits, and it should not be read as one. But it is worth noting that several of the court's framings — particularly its skepticism of the "no market harm" argument given the Times's licensing market for training data — point in directions that will be unhelpful to OpenAI on summary judgment.

Reading the ruling alongside the Second Circuit's Andy Warhol Foundation v. Goldsmith (2023) treatment of transformativeness and the Ninth Circuit's still-recent Authors Guild v. Google framework, the fair-use battlefield as it lands at summary judgment is going to look something like:

• Factor 1 (purpose and character). The Times will argue that ChatGPT's commercial use of Times content for substitutive purposes is the operative use, drawing heavily on Warhol. OpenAI will argue that training is transformative — the model is not "about" the Times articles in any meaningful sense — and will lean on Google Books. Both arguments have force; my prediction is that the court will find a mixed result on factor 1, with the use deemed somewhat transformative but commercial.
• Factor 2. Mostly favors the Times — newspaper articles are at the core of copyright protection — but does not move the needle much.
• Factor 3. The amount used is meaningful. Training uses essentially the entire corpus, but most fair-use cases tolerate substantial copying when the copying is necessary for the transformative use.
• Factor 4. The decisive factor. The Times has a real licensing market for training data — its deal with Amazon is on the docket. OpenAI's position requires arguing that this market is not legally cognizable, which is a hard argument under Warhol.

I would bet on the Times to win on factor 4 and, on the totality, to defeat the fair-use defense. But this is not a confidence-interval-zero prediction. Other district courts have come out differently in adjacent cases.

What this means for the broader docket

There are now more than thirty active U.S. cases involving allegations of copyright infringement through AI training, plus the related but distinct output-infringement cases. The MTD ruling matters across the docket because the framings will be cited heavily.

Specific implications:

1. The discovery-rule holding revives older claims in cases where defendants had pled a 3-year statute of limitations defense. Plaintiffs in Concord Music Group v. Anthropic, Authors Guild v. OpenAI, and several class actions will lean on the SDNY ruling.
2. The DMCA 1202(b) holding is going to be the sleeper expansion of liability. Plaintiffs whose copyright registrations are imperfect or untimely have a parallel statutory hook with no registration prerequisite. Expect more 1202(b) claims to be added in pending cases.
3. The contributory-infringement framing — OpenAI's knowledge of regeneration behaviors as the basis — is a roadmap for plaintiffs to plead around the "neutral platform" defenses that defendants have been mounting. Expect amended complaints across the docket.
4. Settlement pressure on the Times case rises meaningfully. Several other AI labs have signed licensing deals with major publishers in the past year (we've covered them as they happened); the cost-benefit math for OpenAI on a Times deal moves materially after this ruling.

The licensing-market story

One thread worth pulling: the development of the AI training-data licensing market over the last eighteen months changes the legal terrain. When the Times sued in late 2023, the licensing market was largely hypothetical. Today, with deals running into the hundreds of millions and structured products from data providers like Reddit and the Associated Press, the existence of a "potential market" for training-data licenses is no longer hypothetical for purposes of factor 4. This is going to be the most consequential factual development in fair-use analysis.

Bottom line

OpenAI's defense is now harder than it was on the day of filing. The MTD ruling does not decide the case, but it removes the easy off-ramps and pushes the dispute into the fair-use crucible. That is where most observers expected this to land; we are now there. Summary judgment briefing is set for early 2026. Discovery, which has been contentious, will produce its own news between now and then.