Three Fronts, One Week: How September's AI Policy Storm Reshapes Class 42 Brand Strategy
China's top court issued its first AI judicial framework, the US DOJ staked out its position on AI training data, and trademark disputes in New York's busiest IP court jumped 24% - all in September. For any brand operating in Nice Class 42, last month was not business as usual.
A Convergence Brand Owners Cannot Ignore
September 2026 will be cited in trademark seminars for years. Within a single calendar month, three separate jurisdictions moved simultaneously on the legal scaffolding around artificial intelligence, and every movement carries direct consequences for brands that live in Nice Class 42: the home of software, cloud services, and, since 1 January 2026, AI-as-a-Service.
World Trademark Review flagged the convergence in its September AI policy watch, published on 29 September, identifying China, the United States, and the European Commission as the three actors shaping what brand counsel now face heading into Q4.
Advertisement
China Codifies Liability for AI Disputes
The most structurally significant development came from Beijing. On 7 September 2026, China's Supreme People's Court released its Opinions on the Lawful Adjudication of Disputes Involving Artificial Intelligence: the first document issued by China's highest court specifically to guide AI-related disputes. The Opinions run to 24 provisions and span civil liability, intellectual property, consumer protection, autonomous vehicles, and evidence.
For brand owners, the IP provisions are the ones to study. When an AI output infringes copyright, the court divides liability among the developer, the service provider, and the end user. Developers asserting a non-infringement defence may be required to disclose training-data sources, training records, and model operation modes. That evidentiary burden is new, and it sits squarely on the shoulders of companies whose core services are registered under Class 42.
The court did, notably, sidestep the most contested question: whether training AI models on copyrighted works is itself infringing. That fight is deferred. But the framework for everything around it is now codified.
New York's Docket Problem
While Beijing was publishing opinions, the US District Court for the Southern District of New York was quietly becoming the world's busiest venue for AI-adjacent trademark litigation. Exclusive data reported by World Trademark Review shows trademark disputes there jumped 24%, with OpenAI the most-named defendant.
This is not a coincidence of geography. The SDNY's concentration reflects the media and publishing industry's strategic decision to sue where its own companies are domiciled. The September 4 filing by The Seattle Times and Newsday against OpenAI and Microsoft, indexed at 1:2026cv07644, is one of the latest additions to a docket that now includes nearly 400 local and regional newspapers. Several of these suits carry trademark dilution claims alongside copyright arguments, specifically alleging that AI outputs reproducing well-known mastheads dilute the distinctiveness of those marks under federal law.
Class 42 is no longer just where you file your AIaaS mark. It is increasingly where you defend it.
The US DOJ's Quiet Intervention
Also in September, the US Department of Justice filed a statement of interest in the consolidated SDNY litigation, supporting OpenAI's position that barring AI training on copyrighted material could hamper technological progress. The DOJ statement signals that the US executive branch, at least for now, views restrictive training-data rulings as a drag on innovation: a position that distinguishes Washington from Beijing's more liability-focused stance and from Brussels' regulatory instinct.
Brand owners filing or managing portfolios in multiple jurisdictions now face a three-speed world: China with a liability framework but no training-data answer; the US with executive-branch support for AI development but a trial docket exploding with private claims; and the EU still working through its AI Act implementation schedule.
Class 42 Filings: A Warning Signal
Against this backdrop, the UK filing data is arresting. According to AI Business Dispatch analysis of IPO data, Class 42 UK trademark filings stand at just 242 in 2026-Q4 to date, a 97% decline against the equivalent period last year. That figure, sourced from TrademarkDashboard's IPO analysis as of October 2026, is too steep to be seasonal noise.
One reading: applicants are pausing while the jurisdictional picture clarifies. The NCL 13-2026 update, which came into force on 1 January 2026, formally introduced AI as a Service as a standalone term within Class 42, replacing the catch-all use of "software as a service" for AI infrastructure. New filings must now use the updated AIaaS terminology. Some applicants, uncertain whether existing specifications adequately cover the new terminology, may be holding back pending professional review rather than filing speculatively.
That caution may be justified. An overly narrow specification could limit the scope of protection or weaken a mark in non-use cancellation proceedings: exactly the vulnerability that a crowded litigation environment will exploit.
The Naming Gap Is Real
World Trademark Review noted separately this year that AI, Web3, and quantum ventures face a structural naming problem: the speed at which new products reach market consistently outruns the clearance and filing process. The OpenAI "io" injunction, in which a US court blocked use of a product name because it was phonetically identical to a smaller company's registered mark, is the cautionary tale circulating in every in-house team this autumn. Reverse confusion, where a better-funded junior user saturates the market and causes consumers to believe the smaller senior holder is the infringer, is no longer a theoretical risk in the AI sector.
What to Do This Week
If your business provides AI services, whether LLM access, AI consulting, or cloud-delivered machine learning, your Class 42 specification deserves immediate scrutiny. Check whether it references the new AIaaS terminology introduced under NCL 13-2026. If it does not, and if you are operating in or licensing into China, the US, or the EU, the gap between your registered scope and your actual commercial offering is now a litigation risk, not just a filing technicality.
Review via TrademarkDashboard to cross-reference your UK filing position, then take the specification question to a qualified trademark attorney before Q4 ends.