AI and IP Rights: The Trademark Conundrum (Part 2)

By: Sorush Ghodsi (Head of IP Protection)

1. Introduction

In the first installment of this series, we examined how the explosive growth of generative AI is disrupting copyright law, from the legality of training data ingestion to the murky question of AI-generated output. In this second part, we shift our focus to trademark law, another cornerstone of intellectual property protection that is increasingly under stress from AI technologies.

Trademarks serve a function that is distinct from copyright but equally foundational: they allow consumers to identify the commercial origin of goods and services and to distinguish one undertaking’s offerings from another’s. In an era where AI tools can generate logos, brand names, slogans, and even sounds within seconds, and where AI-driven advertising systems can systematically exploit established marks at an industrial scale, the traditional trademark framework faces challenges that are only beginning to be fully understood.

1. The Legal Framework: What Qualifies as a Trademark?

Before turning to the AI-specific challenges, it is necessary to outline the EU trademark framework briefly. The primary legislative instruments are the EU Trade Mark Regulation (EUTMR), Regulation (EU) 2017/1001, which governs EU Trade Marks (EUTMs) administered by the EUIPO, and the Trade Mark Directive (TMD), Directive (EU) 2015/2436, which harmonizes national trademark laws across Member States.

1. Signs and the Representation Requirement

Article 4 of the EUTMR defines what can constitute a trademark: any sign capable of distinguishing the goods or services of one undertaking from those of others, provided that it can be represented on the Register in a way that enables the competent authorities and the public to determine the clear and precise subject matter of protection. This representation requirement is not trivial. In the landmark Sieckmann case (C-273/00)[1], the CJEU held that a sign must satisfy seven cumulative criteria: it must be clear, precise, self-contained, easily accessible, intelligible, durable, and objective. While this standard was developed in the context of olfactory marks, it has since been applied more broadly to non-conventional signs such as sounds, holograms, and motion marks.

The EU framework is therefore technologically neutral as to the type of sign, words, logos, shapes, colours, sounds, and even certain non-visual signs can all attract trademark protection, but it is strict about the quality of representation.

• Absolute Grounds for Refusal

Even where a sign satisfies the representation requirement, registration may be refused on absolute grounds under Article 7(1) EUTMR. The most significant are:

• Lack of distinctive character (Article 7(1)(b)): the sign must be capable of identifying the commercial origin of the goods or services in the mind of the average consumer. Purely generic or commonplace signs fail at this hurdle.
• Descriptiveness (Article 7(1)(c)): signs that merely designate the kind, quality, quantity, intended purpose, value, geographical origin, or other characteristics of the goods or services are not registrable without acquired distinctiveness through use (secondary meaning).
• Genericness (Article 7(1)(d)): signs that have become customary in trade, the common name for a product in the language of the relevant public, are excluded from protection.
• Functional shapes (Article 7(1)(e)): signs consisting exclusively of the shape (or other characteristic) of goods that is necessary to obtain a technical result, that results from the nature of the goods themselves, or that gives substantial value to the goods, are not registrable.
• Deceptiveness (Article 7(1)(g)): signs of a nature to deceive the public, for instance as to the nature, quality, or geographical origin of the goods or services, are refused registration.

These grounds are assessed ex officio by the EUIPO during examination and reflect a public interest in keeping certain signs freely available for all economic operators.

•  Relative Grounds: Likelihood of Confusion

Beyond absolute grounds, registration may be opposed, and an existing trademark infringed on relative grounds. The most invoked basis is the likelihood of confusion under Article 8(1)(b) EUTMR (opposition) and Article 9(2)(b) (infringement). The test requires a global appreciation of all relevant factors, including the visual, phonetic, and conceptual similarity between the signs, the similarity of the designated goods or services, and the distinctiveness of the earlier mark. The CJEU established the analytical framework in Sabel v Puma (C-251/95) and Canon v Metro-Goldwyn-Mayer (C-39/97), confirming that a lesser degree of similarity between the goods or services may be offset by a greater degree of similarity between the marks, and vice versa.

1. Registration Pathways: National Filing, the Madrid Protocol, and the EUTM

For businesses seeking trademark protection, the EU offers two main pathways, in addition to the international route.

The first is national registration with the IP office of each individual Member State. This approach offers precise territorial control but is administratively intensive for rights holders targeting multiple markets. Once a national registration is obtained in a Member State that is party to the Madrid System, the proprietor may use that registration as a “home” base to extend protection to additional contracting states through the Madrid Protocol, administered by WIPO.[2] With more than 130 contracting parties, the Madrid System is the primary mechanism for securing multi-jurisdictional trademark protection from a single filing, making it particularly attractive for IP-intensive businesses expanding globally.

The second option, generally preferred for EU market strategies, is the EU Trade Mark (EUTM), applied for directly with the EUIPO. A single EUTM grants protection simultaneously across all 27 EU Member States at a substantially lower cost per jurisdiction than multiple national filings. The EUTM is unitary in nature; it cannot be divided territorially, meaning a successful invalidity or revocation action affects the entire EU territory, a consequence that demands careful clearance strategy before filing.

• AI-Specific Challenges

Having established the foundational framework, we can now examine how AI technologies intersect with and challenge each of its key components.

2.1. The Registrability Question: Can an AI-Generated Sign Be a Trademark?

One of the most immediate questions raised by generative AI in the trademark context is whether a sign produced by an AI tool, a logo, a sound, a slogan can be validly registered as a trademark.

Structurally, this question differs significantly from the copyright equivalent examined in Part 1. Copyright protection, as interpreted by the CJEU in Infopaq and its progeny, is conditioned on the existence of a human “author’s own intellectual creation.” Trademark law imposes no equivalent authorship requirement. The criterion for registration is functional and commercial, not authorial: can the sign distinguish? Is it capable of representation? Does it fall within any of the absolute grounds for refusal? The tool or process by which the sign was created is irrelevant to these assessments under the current EUTMR framework.

In practice, neither the EUIPO nor national offices systematically inquire into the creative process behind a submitted sign. Examination is largely formal, assessing distinctiveness, representation quality, and absolute grounds, and any challenge to registrability grounded in AI generation would only arise in opposition or invalidity proceedings, for which the EUTMR provides no explicit basis.

Two practical complications nonetheless deserve attention. First, who is the legal applicant? Article 5 EUTMR requires that a trademark be applied for by a natural or legal person. An AI system cannot hold legal personality and therefore cannot be an applicant; ownership must vest in the human or legal entity that deployed the tool. Second, and more subtly, certain AI platform terms of service limit the exclusivity of rights transferred to the user over generated outputs. Where a company cannot demonstrate full exclusive control over a sign, for instance, because a platform retains residual rights or licenses the same output to multiple users, the lawfulness requirement under Article 7(1)(g) EUTMR may be engaged, potentially rendering the mark unregistrable or vulnerable to invalidity. Practitioners advising clients on AI-assisted branding must therefore scrutinize platform licensing terms before filing.

Subject to these caveats, an AI-generated sign that satisfies all substantive requirements appears registrable under current EU law.

2.2. The Infringing Output Problem: AI-Generated Look-Alike Signs

While the registrability of AI-generated marks is a relatively contained doctrinal question, the output side of the equation raises far more serious concerns for existing trademark proprietors.

Generative AI systems trained on large datasets, which inevitably include the full corpus of registered and unregistered trademarks in commercial use, are capable of producing signs that are confusingly similar to pre-existing protected marks. This can occur through deliberate prompting (“generate a logo in the style of Brand X”) but also unintentionally, as a statistical artifact of training on brand-heavy image datasets. AI models optimizing for visual plausibility may converge on patterns that are statistically “brand-like” precisely because established marks dominate their training distribution.

Under the CJEU’s global appreciation framework, a likelihood of confusion can arise even where the similarity between two signs is not perfect. The average consumer’s overall impression, assessed visually, phonetically, and conceptually, is determinative. A company that adopts an AI-generated logo without conducting a proper clearance search risks infringing rights it was entirely unaware of. This matters because, under EU trademark law, infringement does not require intent or knowledge of the prior right: the existence of a likelihood of confusion is sufficient for liability.

This creates a structural risk peculiar to AI-assisted branding. Human-led design processes involve a degree of self-limiting originality; designers actively seek to differentiate themselves from competitors as a professional matter. AI generative models, operating to a different objective function, may produce outputs that cluster around familiar commercial visual patterns without any corrective mechanism analogous to professional judgment.

2.3. Liability Allocation: Developer, User, or Platform?

The question of who bears legal responsibility when an AI tool generates an infringing sign is one of the most legally uncertain areas at the intersection of AI and trademark law.

Under existing EU doctrine, three actors are potentially in the frame.

• The end user is the most straightforward candidate for primary liability under the existing doctrine. A company that adopts an AI-generated mark in its commercial activities without adequate clearance uses the sign in the course of trade within the meaning of Article 9 EUTMR and bears direct infringement liability regardless of how the sign was created.
• The AI developer presents a harder case. EU trademark law, unlike some common law systems, does not recognize a general theory of contributory or secondary trademark infringement. As Harvard Law Review noted in its commentary on Google France, the concept of “contributory infringement” is largely “foreign” to EU trademark doctrine. Nevertheless, developers who design AI systems with a foreseeable capacity to generate confusingly similar outputs may face non-TM claims in product liability or tortious negligence frameworks, a question that remains unresolved in EU case law.
• The platform question turns on intermediary liability. The Digital Services Act maintains the conditional safe harbor from the E-Commerce Directive: hosting providers are shielded from liability for third-party content they neither knew about nor had control over, provided they act expeditiously to remove infringing content upon obtaining actual knowledge. However, the safe harbor is conditional on “passive” conduct. The CJEU in Google France (C-236/08 to C-238/08)[3] made clear that a service provider that plays an active role, for instance by optimizing, selecting, or shaping content, may lose the benefit of the exemption. A generative AI platform that actively shapes its outputs (rather than merely hosting user-submitted content) may be placed in precisely this exposed position. The DSA’s new due diligence obligations for very large online platforms, including notice-and-action mechanisms and mandatory risk assessments, add a further layer of compliance pressure relevant to AI-generated brand content.

2.4. AI in Keyword Advertising: A New Dimension of an Old Problem

The intersection of AI and keyword advertising represents perhaps the most commercially significant trademark challenge in the current digital economy. The foundational EU framework was established in the joined cases Google France (C-236/08 to C-238/08), where the CJEU held that Google did not infringe trademarks by allowing advertisers to purchase keywords corresponding to protected marks, so long as Google’s role remained purely technical, automatic, and passive. However, the Court confirmed that advertisers who bid on competitors’ marks as keywords may infringe those marks where the resulting advertisements do not allow the reasonably well-informed internet user to determine the commercial origin of the advertised goods or services. The CJEU refined these principles in Interflora v Marks & Spencer (C-323/09), particularly on the question of dilution of well-known marks through keyword exploitation.

AI has materially altered this landscape in two structurally important ways. First, AI-powered bidding systems, now standard across major digital advertising platforms, autonomously identify, select, and bid on keywords in real time, including competitors’ brand names, without explicit human instruction for each bid. The question of whether this degree of automation changes the active/passive characterization central to Google France is not yet settled in EU case law, but it is a live and commercially urgent question. Second, large language model-driven advertising copy generators can produce text that is deliberately ambiguous as to commercial origin, exploiting the consumer’s inability to identify the advertiser’s relationship to the referenced brand.

Senftleben’s proposal for parameter transparency obligations for AI advertising systems, requiring disclosure of the algorithmic factors governing keyword selection and ad targeting, is directly responsive to this challenge and aligns with the spirit of the DSA’s recommendation system transparency requirements already applicable to VLOPs.

2.5. AI-Powered Counterfeiting, Domain Spoofing, and Algorithmic Manipulation

Beyond registered trademark rights, AI technologies present a growing and underappreciated threat to brand integrity at the operational enforcement level. Three risks merit specific attention.

1. Counterfeit product imagery. Generative image AI can produce high-quality packaging and product imagery that closely mimics established brands, dramatically reducing the skill and capital required to mount a convincing counterfeiting operation. The resulting enforcement challenge is asymmetric: identifying and challenging individual instances of counterfeit imagery at scale demands significant resources from brand owners, while the marginal cost of production for infringers is negligible. The DSA’s obligations on online marketplaces, including notice-and-action systems and proactive obligations for VLOPs, provide the most immediate regulatory lever for platforming of AI-generated counterfeit content.
2. Domain spoofing and lookalike websites. AI tools can generate convincing lookalike websites, replicating the visual identity, layout, and product descriptions of established e-commerce operators, and can produce infringing domain names that exploit typographical or semantic proximity to protected marks. Such operations engage both trademark infringement under the EUTMR and unfair commercial practices under Directive 2005/29/EC. WIPO’s Uniform Domain-Name Dispute-Resolution Policy remains the standard enforcement mechanism for infringing domain names, but its procedural timeline is ill-calibrated for the speed at which AI-enabled spoofing infrastructure can be deployed and rotated.
3. Algorithmic manipulation and search ranking. Perhaps most insidiously, AI-optimised SEO and automated pay-per-click strategies can exploit the brand recognition of established marks to divert consumer traffic to counterfeit or competing products, causing dilution and tarnishment without direct copying of the sign. These practices operate beneath the threshold of traditional infringement doctrine; no sign is reproduced, no confusion in the classical sense arises, yet their commercial effect on brand equity is real and quantifiable. The DSA’s audit obligations for very large platforms offer the most promising near-term instrument for addressing systemic manipulation of this kind.

Conclusion: A Framework Under Pressure

EU trademark law was designed for a world in which brand creation, exploitation, and infringement occurred at human speed and scale. Generative AI disrupts each of these assumptions simultaneously. Signs can now be created at machine speed; infringement can propagate across platforms in real time; liability is diffused across chains of automated actors; and enforcement tools calibrated for individual human infringers are under severe strain.

Some of these challenges are manageable within the existing doctrine. The registrability of AI-generated signs, the primary liability of the end user for infringing output, and the basic framework of likelihood of confusion analysis remain operable under current rules. But the gaps in liability allocation between developers and platforms, the stress placed on the Google France framework by autonomous AI bidding systems, and the scale of AI-enabled counterfeiting all point toward the need for legislative or regulatory clarification.

The EU AI Act offers a partial but incomplete response. Its transparency obligations under Article 50, requiring providers of general-purpose AI models to disclose training data summaries and respect copyright opt-outs, have indirect relevance to trademark-laden training datasets, but the Act contains no trademark-specific provisions. More pertinently, AI systems used to generate brand identities or automate advertising decisions are unlikely to meet the threshold for classification as high-risk systems under Annex III, meaning the Act’s most demanding conformity assessment and human oversight obligations will not apply to the majority of AI-driven brand tools currently in commercial use. The practical implication is that the AI Act’s implementing framework, as it stands, does not fill the enforcement gaps identified in this article. Targeted amendment of the EUTMR or sector-specific DSA obligations tailored to brand protection remains the more promising legislative lever.

What is clear is that traditional trademark enforcement tools, clearance searches, opposition proceedings, UDRP filings, and infringement actions must be supplemented with a more sophisticated understanding of how AI systems generate, deploy, and exploit signs in the commercial sphere. Brand owners, practitioners, and regulators who grasp this dynamic early will be better positioned to protect the foundational function that trademark law has always served: ensuring that consumers can trust the origin of what they purchase.

Bibliography

Regulation (EU) 2017/1001 of the European Parliament and of the Council of 14 June 2017 on the European Union trade mark [2017] OJ L154/1

Directive (EU) 2015/2436 of the European Parliament and of the Council of 16 December 2015 to approximate the laws of the Member States relating to trade marks [2015] OJ L336/1

Regulation (EU) 2022/2065 of the European Parliament and of the Council of 19 October 2022 on a Single Market For Digital Services (Digital Services Act) [2022] OJ L277/1

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L2024/1689

Regulation (EU) 2024/1689 of the European Parliament and of the Council of 13 June 2024 laying down harmonised rules on artificial intelligence (Artificial Intelligence Act) [2024] OJ L2024/1689

Directive 2005/29/EC of the European Parliament and of the Council of 11 May 2005 concerning unfair business-to-consumer commercial practices [2005] OJ L149/22

Case C-273/00 Ralf Sieckmann v Deutsches Patent- und Markenamt [2002] ECR I-11737

Case C-251/95 SABEL BV v Puma AG, Rudolf Dassler Sport [1997] ECR I-6191

Case C-39/97 Canon Kabushiki Kaisha v Metro-Goldwyn-Mayer Inc [1998] ECR I-5507

Joined Cases C-236/08 to C-238/08 Google France SARL and Google Inc v Louis Vuitton Malletier SA and others [2010] ECR I-2417

Case C-323/09 Interflora Inc v Marks & Spencer plc [2011] ECR I-8625

Martin Senftleben, ‘Trademark Law, AI-Driven Behavioral Advertising, and the Digital Services Act: Toward Source and Parameter Transparency for Consumers, Brand Owners, and Competitors’ in Ryan Abbott (ed), Research Handbook on Intellectual Property and Artificial Intelligence (Edward Elgar 2022) 309

Melissa Biesmann, ‘AI, the New Frontier: An Analysis on Trademark Litigation Strategies in the Face of Generative Artificial Intelligence’ (2024) 29(1) Marquette Intellectual Property & Innovation Law Review 33

Louise Curtis and Rachel Platts, ‘Trademark Law Playing Catch-Up with Artificial Intelligence?’ (WIPO Magazine, June 2020)

‘Artificial Intelligence and Trademark Infringement: Legal Interpretation of “Use” of AI-Generated Signs in the European Union’ (2026) Trends in Intellectual Property Research https://iprtrends.com/TIPR/article/view/98

---

[1] https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=celex:62000CJ0273

[2] https://madrid.wipo.int/

[3] https://infocuria.curia.europa.eu/tabs/redirect/juris/liste.jsf?num=C-236/08&language=en