Invest1 distinct publisher2 min readPublished
The 24th US copyright case against OpenAI wants a court order barring scraped how-to articles from future training runs. Only about a tenth of the articles carry registrations.
The Investor · Invest desk

Compiled by The InvestorSomething wrong?How this is made
The remedy request is the part of this filing that does work a cheque cannot. WikiHow is asking the court to order OpenAI to stop using the scraped material, not merely to pay for having used it [6], and the complaint's framing is that such an order could require retraining affected models [7]. Damages get priced, budgeted and absorbed. An order that reaches forward into training runs attaches to the pipeline itself, and it binds conduct that has not happened yet. That is why remedy scope, and not just liability, is worth tracking in every one of these dockets.
The registration count sets up the other half of the case. At least 1,200 of the articles carry registered copyrights [3], against a claimed corpus of more than 11,000 scraped articles [1]. That is under 11 percent [12]. Registration is what unlocks statutory damages in US court, a sharper instrument than actual damages [5], so roughly 9,800 articles ride on WikiHow proving real economic harm [13]. Its argument there is displacement: ChatGPT now produces step-by-step instructional content that competes with its articles, faster and at a fraction of the production cost [8], which the complaint ties to lost traffic and advertising revenue [9]. Proving that means WikiHow's own analytics become evidence.
The plaintiff profile matters for reasons that have nothing to do with sympathy. WikiHow is neither the New York Times, which brought its own suit [10], nor a lone author [11]. Its library is large, uniform in format and heavily indexed by search engines, which made it useful as training material and also makes the alleged copying comparatively cheap to document [14]. Evidentiary cost is the thing that usually decides whether a mid-sized publisher can afford to litigate at all.
OpenAI's answer is the one it has given in effectively every comparable case: the models were trained on publicly available data, and that use is fair use [15]. The case sits in the Southern District of New York alongside several other AI copyright suits [4][16], where the fair use questions in play are the commercial character of the use and the effect on the market for the original [17]. Instructional writing is an awkward place for a defendant on that second point, because a numbered list substitutes for a numbered list.
Ranked by verification strength, evidence, and original report placement.
The complaint was filed on August 21, 2026, in the US District Court for the Southern District of New York.
The WikiHow filing is the 24th copyright suit brought against OpenAI in US courts; OpenAI is simultaneously defending claims from news publishers, book authors, visual artists and now a wiki-style instructional platform.
The lawsuit argues that if AI models trained on WikiHow content replicate its output on demand, the site loses traffic and advertising revenue, reducing the incentive to keep producing the original content.
OpenAI maintains that its models are trained on publicly available data and that such use falls within fair use, the response it has used in virtually every similar case.
The Southern District of New York is the same court handling several other high-profile AI copyright suits.
WikiHow filed a copyright infringement lawsuit against OpenAI claiming the company scraped more than 11,000 of its instructional articles to train GPT models, including ChatGPT, without permission.
Follow any of these and your For You feed starts watching them — no settings page required.
Evidence-backed comparisons of source perspectives and observed adoption signals. Read the methodology
Which Builder, Operator, and Investor concerns the observed source mix emphasized—not a truth score.
Evidence, demonstrated adoption, hype gap, incentives, and confidence are assessed independently, each on its own current evidence. How these are measured.
Single aggregated source, no primary filing
One publisher, itself crediting gizmodo.com, carries every fact in the cluster. There is no docket number, case caption or link to the complaint, no independent legal reporting, and no case-specific statement from OpenAI. The internal figures are at least self-consistent and the described corpus characteristics are checkable, which keeps this above the floor, but nothing here is corroborated.
Not an adoption story
The only dated event in the cluster is a court filing. There are no releases, deployments, benchmarks, pricing or licence changes, or usage disclosures, and no traffic or revenue figures for WikiHow, so there is nothing to measure as adoption. Inferring uptake from litigation activity would be a guess.
Headline scale outruns recoverable exposure
The 11,000-article figure carries the headline while the statutory-damages hook attaches to at most 1,200 registered works, under 11 percent of the total; the remainder depends on proving actual damages that the article never quantifies. The retraining scenario is presented as a live consequence though it is conditional on relief no court has granted. The overstatement is modest rather than severe because the piece itself discloses the registration subset, names no damages figure, and hedges the injunction discussion.
Litigant motives visible on both sides
Both parties' incentives are legible from the cluster itself. WikiHow's own filing frames the case as a financial survival argument about lost traffic and advertising revenue, and the source notes its corpus is unusually easy to document as training input. OpenAI's fair use posture is described as the response it deploys in virtually every similar case, i.e. a standing defensive interest across 24 suits. The publisher's own incentive is less visible: it is a crypto outlet republishing a Gizmodo report with no disclosed relationship to either party.
Plausible but uncorroborated
The core facts are the kind that are easy to verify and hard to fabricate, and the internal arithmetic holds. But with one aggregating publisher, no primary document, no docket identifier, no OpenAI comment and no second outlet, confidence cannot rise far. The forecast element about retraining is the weakest link and is marked insufficient on its own.
invest
OpenAI ships ChatGPT ads in six markets with a two-tier consent regime1 distinct publisher
product
A school agenda shipped with "Vitoiis" and a planet named Marc, and no one read it first1 distinct publisher
invest
Storm's reductio: if the DOJ theory catches him, it catches OpenAI and Google1 distinct publisher
product
Alsup fined the download, not the training: the copyright risk in your stack is provenance1 distinct publisher
Distinct publishers with included, body-backed reporting in this cluster.
cryptobriefing.com
1 article · August 24, 2026