The Name That Describes Itself: What OpenAI's EU Trademark Defeat Means for Every AI Brand

Fame Is Not Distinctiveness

There is an old trap in trade mark law: confusing how well-known a name is with how protectable it is. OpenAI walked into it on 15 July 2026, when the General Court of the European Union handed down its judgment in Case T-555/25, OpenAI, Inc. v EUIPO. The court confirmed the EUIPO's partial refusal to register the word mark OPENAI for goods and services in Classes 9, 42 and 45, covering software, cloud-computing services, and identity verification.

The reasoning is brisk and uncomfortable. The EUIPO found that, for at least a non-negligible proportion of the English-speaking public, OPENAI immediately conveys that the goods and services are provided using, or connected with, openly accessible artificial intelligence: "open" AI, abbreviated as "AI". The General Court agreed. Hundreds of millions of users recognise the name; that recognition, the court held, does not cure the underlying descriptiveness. Reputation does not affect the intrinsic assessment.

One route remains open. OpenAI can still attempt to establish acquired distinctiveness under Article 7(3) of the EU Trade Mark Regulation, by demonstrating that consumers associate OPENAI specifically with one commercial source. The ruling can also be appealed to the Court of Justice of the European Union. But the primary path, inherent distinctiveness, is gone for the contested goods and services. The OpenAI logo, as a figurative mark, remains registered. It is the words alone that are the problem.

The Wider Signal for AI Branding

This case matters well beyond one San Francisco company. The judgment shows how difficult it is to secure an exclusive right to designations built from the technical vocabulary of AI, even where those designations have become commercially dominant. Any brand that appended "AI", "GPT", "Gen", "Neural", or similar descriptors to a generic word and hoped that would be enough should treat T-555/25 as a warning.

The 13th edition of the Nice Classification, in force since 1 January 2026, has formalised "artificial intelligence as a service" (AIaaS) as an explicit entry in Class 42. That is a helpful clarification for specification drafting, but it also concentrates examiner attention: the more precisely regulators define what AI services are, the easier it becomes to argue that a mark merely describes them. Brands filing in Class 42 for AI-delivered services, or in Class 9 for AI-enabled software, now face an environment where offices on both sides of the Channel are actively sharpening their descriptiveness tests.

At the EUIPO, the 2026 Guidelines, in force from 1 July, require precision, evidence and consistency throughout specification drafting. Generic or strategically vague wording is no longer acceptable: virtual goods must be specified clearly, and "virtual goods" alone is refused unless qualified, for example as "virtual goods, namely virtual clothing".

The UKIPO is preparing its own shift. According to Stevens & Bolton, writing on 14 July 2026, the UKIPO will begin a discovery phase for updating its trade mark search and filing platform from September 2026, as part of the One IPO Transformation Programme. The office has already identified AI as a priority within this programme, responding in part to a surge in non-traditional filings, including sound marks, multimedia marks, and personality and image-rights applications, driven partly by concern about AI-generated content replicating image and likeness.

The Class 36 Blind Spot in Legal Services

There is a related vulnerability in the UK data. AI Business Dispatch analysis of Companies House and IPO data (as of July 2026) shows that UK trade mark filings in Class 36, financial services, reached only 660 in Q3 2026, a 72.3% fall versus the prior period. Alongside that, just 153 new SIC 69.10 companies (activities auxiliary to financial services) were incorporated in Q3 2026, down 78.3% year-on-year. Most striking: 98.8% of active SIC 69.10 companies hold no Class 36 trade mark at all.

Class 36 may not be the obvious home for an AI brand, but the pattern it reveals is consistent across the economy. UK companies operating in AI-adjacent sectors are incorporating and trading without securing any registered trade mark protection. When a competitor, a counterfeiter, or an AI-generated brand name occupies a similar name in the relevant class, those companies have no registered right to oppose with.

The Clarivate Trademark Filing Trends Report 2026 makes the same point from the filing side: despite rapid commercial influence, most major AI companies did not feature among the highest trade mark filers, suggesting brand formalisation in the sector is still catching up.

What Brand Owners Should Do This Week

The lesson from Luxembourg is specific. If your brand name can be parsed as a description of what AI does, a word-mark application alone is not a safe strategy. Pair it with a distinctive figurative mark that can be registered independently. Commission a clearance search that checks not only for identical or similar earlier marks but also for descriptiveness risk under the relevant class headings: Class 9 for software, Class 42 for AI-as-a-service, Class 45 for identity and legal services.

UK brand owners can check their Class 42 and Class 9 exposure against the live register at TrademarkDashboard before committing to a name or a filing strategy. The UKIPO's September discovery phase may bring changes to how non-traditional marks are searched, so any portfolio review begun now will be ahead of that curve.

If you are running an AI business under a name that describes AI, take advice from a qualified trade mark attorney before the next filing cycle. The EU General Court has just illustrated, at considerable cost to one of the world's best-known technology companies, that fame and registrability are not the same thing.