Trademark Liability For Automated Brand - Name Generators
Introduction : Naming a brand has entered an era in which one types a product description into a generative algorithm and sees an instantly produced list of pithy, available-sounding names to choose from as the first step in the launch process. What one types into the algorithm and the list it produces have little bearing on the ability of the selected name to withstand claims of trademark infringement: whether the name one chose appears among hundreds of unregistered marks used in India in the relevant space; whether the name in another, linguistically similar language was already registered in India or acquired goodwill; and whether the use of the chosen name is an infringement, even without bad faith on the part of the user, all remain open questions. This article addresses those issues as they arise in the context of a brand name selected, at least in part, using AI-generated suggestions, focusing on the ability of trademarks law to address the unintentional reproduction of existing marks, the responsibility for such clearance searches, the differences between phonetic, visual, and conceptual similarity, the risks posed by unregistered or common-law marks, the liability, if any, of the AI’s creator, and concluding with a suggested pre-launch naming protocol.
Legal Provisions
Section 29, Trade Marks Act , 1999. Trademark infringement occurs when a person, not being the registered proprietor of a trademark, uses in the course of trade “a mark identical with, or deceptively similar to, the registered trademark for the same description of goods or services as those in respect of which the trademark is registered,” thereby causing confusion or deception about the origin of the goods or services. Section 29 is a civil wrong with near-strict liability, meaning that the infringer’s knowledge or intent is not a factor in determining liability. In other words, it does not matter if the infringer used the AI tool in good faith or was unaware of the pre-existing mark; they are still secondarily liable for trademark dilution (or, in the case of unregistered, common-law marks, passing off). Section 29(4), in turn, provides the same degree of protection as would otherwise be afforded by virtue of the passing off action to registered trademarks, against a defendant’s using them in relation to goods/services not of the plaintiff’s making, but in a manner that takes “unfair advantage of, or detrimental to, the distinctive character of the mark” - including using similar marks as applied to an entirely unrelated product.
Section 27 and passing off, in cases of passing off, in which a trademark was unregistered or became unregistered or failed to register, the common law tort of passing off provides relief: essentially the same protections as a registered trademark offers to the plaintiff, even if the plaintiff lacks registration or the infringing party was unaware of the original trademark’s existence. Passing off is a high bar to prove, but it is not impossible for an AI-generated name to fall afoul of, particularly in the case of an unregistered, common-law mark with existing reputation. As with infringement, passing off does not require proof of “consciousness of the plaintiff’s fame,” meaning that an infringer is liable even if they were simply unaware of the existing name, trademark, or brand.
Section 103 and criminal liability. Unlike the civil wrong of infringement under the Trade Marks Act, the penal provision of Section 103 makes “falsely applying for registering trademarks” a punishable offense, with punishment ranging between a fine and imprisonment for up to 2 years. However, the provision uses the same standard of proof as Section 27: a defendant is only criminally liable for trademark infringement if they were knowingly or “had reason to believe” that such use of the trademark would be deceptive, not if they acted in good faith or were unaware of existing marks.
The Cadila deceptive-similarity framework
In the case of Cadila Health Care Ltd v. Cadila Pharmaceuticals Ltd. , (2001) 5 SCC 73, the Supreme Court held that “deceptive similarity” between trademarks is a test that considers a number of individual factors, including the nature of the marks, the extent of similarity in appearance, sound, and meaning, the nature of the goods or services, the similarities between them, the class of purchasers, and the manner of purchase. The court also held that “in a country like India, where there is a great diversity of languages and varying degrees of education among the populace,” “a court should be especially careful when marks are merely phonetically similar” and attach more weight to the phonetic test than it might elsewhere. As such, even if the infringing mark differs significantly from the plaintiff’s trademark, such similarity in sound or structure is enough to incur liability. The framework applies to comparisons between AI-generated names and existing marks, even if the former were created by an AI, because it is designed to evaluate the similarity of two human-created trademarks.
Domain name frameworks
INDRP and UDRP. India does not have a dedicated “cybersquatting” statute; instead, domain name disputes are evaluated under the same passing-off doctrine as trademark infringement or by the IN Dispute Resolution Policy (INDRP), a policy on dispute resolution in domain name registrations administered by the National Internet Exchange of India and modelled on the Uniform Domain-Name Dispute-Resolution Policy (UDRP) of ICANN/WIPO. Both INDRP and UDRP require that in order to bring a complaint against a domain name registrant for trademark infringement, the complainant must prove that (i) the domain name is “identical or confusingly similar” to their trademark, (ii) the respondent has no legitimate right or interest in the domain name, and (iii) the respondent registered or used the domain name “in bad faith.”
IT Act intermediary status of AI providers. Under the Information Technology Act 2000 and the rules governing “intermediaries,” MeitY has noted that an AI tool is a service that provides information in response to queries from other users; in this capacity, it occupies the “passive conduit” position and, therefore, qualifies for safe harbour under Section 79 of the IT Act. This means that an AI tool, in its default capacity as a passive conduit, is not responsible for illegal content published through its system and is only obliged to remove or disable access to such content upon notice. However, a tool that “takes active steps for the publication or transmission of information,” on the other hand, does not qualify for safe harbour and can be held secondarily liable for user-generated content in breach of the IT Act’s rules. As such, naming tools that allow users to publish content or names generated by the AI may find themselves secondarily liable for trademark infringement if they know or should have known of the illegality of the content but failed to act accordingly.
Legal Analysis
How collisions occur. A generative tool learns patterns – phonetic patterns in particular, since brand names are overwhelmingly sound-based – and uses that knowledge to generate new names. This inherently makes generative tools prone to name collisions, since the AI has learned from a corpus of names that appear similar to one another, thus prioritizing phonetic patterns that already exist. In other words, independently generated names will tend to cluster around sets of similar-sounding existing names and, as such, have a disproportionately high risk of infringing on an existing trademark. Furthermore, the tool has no access to the full breadth of trademarks, unregistered marks, or common-law brands, and may suggest a name that is phonetically identical to an existing one but spelled differently or has regional variations in other languages.
Search responsibility
No party can treat the use of a generative tool as a substitute for a comprehensive trademark search. Even if the search is conducted by the branding agency, the responsibility for ensuring that the shortlisted names have been cleared of trademark infringement rests with the client, who is the end-user of the brand name and who can be held secondarily liable for trademark infringement by virtue of using the mark in the course of trade. This makes the division of responsibility between the client, agency, and AI unclear, but the clearest resolution is for the client explicitly to request a trademark clearance search prior to the agency shortlisting potential names for consideration.
Similarity analysis for AI-generated names
Similarity testing, including the application of the Cadila factors, applies to AI-generated names in the same way as to two traditional trademarks. Two additions qualify AI-generated names for consideration under the framework: first, invented words (which the tool may generate) undergo analysis for phonetic similarity, since there is no lexical meaning to indicate conceptual similarity; and second, because generative AI tools are often used to propose variations of a root name, one must treat each similar-sounding variation as a potentially infringing mark.
Domain availability as a distinct, cascading risk
Domain names are often confused with trademarks, but the two are distinct, and a domain name’s availability does not necessarily indicate a trademark’s availability (or the converse). The same word or its variations may be trademark- and domain-clear but still fall afoul of UDRP/INDRP in bad-faith squatting disputes after the trademark attains a certain degree of recognition and the domain owner tries to profit from that by selling it at an inflated price. Furthermore, because domain names prioritize phonetic similarity, names that would not infringe a trademark (even one with similar sounds) can still be ruled confusingly similar under UDRP/INDRP and subject to a dispute. As such, a domain name’s availability should be treated as a separate consideration from trademark infringement – one that runs concurrently with it rather than being contingent on trademark clearance.
Linguistic conflicts in the Indian context
The Indian context is relevant to the phonetic analysis of both AI- and manually generated trademarks under the Cadila framework. Cadila holds that in analyzing deceptive similarity, the court must take into account “the diversity of languages and varying degree of education among the populace,” and so should apply weight to phonetic similarity, even if the resemblance is limited to the manner in which the word sounds. A generative tool, which learns and applies patterns in sound, is likely to select names that are phonetically similar to existing names in ways that may conflict with trademarks or brands, especially since most tools are trained primarily on English, which may lead to a false sense of security with regard to other Indian languages. As such, an additional layer of scrutiny for phonetic similarity in other Indian languages is appropriate for names suggested by generative tools.
Allocation of Risk Between Client, Agency and AI Provider
The client bears ultimate statutory liability. Under Indian law, trademark infringement is a strict civil liability, and the infringing party, when using a trademark in the course of trade, is secondarily liable for damages suffered by the trademark proprietor due to the unauthorized use. This does not differentiate between trademarks deliberately infringed and trademarks accidentally infringed by virtue of using a name suggested by a generative AI tool. The client, therefore, is the party that bears the ultimate legal liability for trademark infringement. That said, a contractual obligation between the client and the agency, in which the latter indemnifies the former, absolves the agency of any further liability, and the parties to the contract negotiate how the risk is allocated.
The branding agency bears the risk of professional negligence
A branding agency signs a contract with the client to deliver a branding proposal with a trademark suggestion and/or shortlisting, and so the agency bears the risk of professional negligence inherent to the job. The agency is expected to possess a reasonable level of skill and competence as a contractor in its field and to adhere to professional standards of conduct, including conducting a search to ensure that the trademark has been cleared for registration. Thus, if the agency fails to do so, it may be found liable for professional negligence for misadvising the client in the selection of a trademark, even if the agency itself was not infringing any third party’s trademark.
The AI provider bears potential liability for failure to meet standards of notice
An AI provider, in turn, occupies an ambiguous position in the professional responsibility chain, due to the absence of detailed regulations for generative AI tools in India. Most such tools do not carry any responsibility for the content suggested by the AI beyond clearly disclaiming any professional responsibility in their terms of service. Furthermore, under the IT Act’s rules on “intermediaries,” tools that operate as “passive conduits” are granted safe harbour from liability but are responsible for promptly removing content in response to a takedown notice, meaning that an AI provider that was not aware of an infringement but had received a takedown notice would be liable for failure to act on it. On the other hand, an AI tool that functions as a “publisher or a service provider” has greater obligations and may be held responsible for content it knew or should have known was infringing. As such, a trademark infringement committed using an AI tool does not automatically render the tool’s publisher liable, but may do so if the tool takes a position as a publisher, knowingly permits infringement by users, or fails to remove infringing content upon receiving an official notice.
Practical Implications
For clients, the key takeaway is that AI-assisted brand name selection shortens the timeline for idea generation but does not shorten the timeline for trademark clearance – an AI-generated shortlist carries the same risks and considerations as a human-suggested name. For branding agencies, the most obvious takeaway is that the disclosure of a tool’s use and the allocation of responsibility for clearance in the engagement letter are integral risk management measures and serve as a contractual shield against any accusations of professional negligence. For AI providers, the liability considerations cut both ways: the push to re-brand AI tools as “end-to-end branding solutions” competes with the practical difficulty of ensuring that no suggested name infringes any existing trademark, with liability implications for the provider in the event of a dispute.
Conclusion
Brand name generators do not create new difficulties for trademarks law; rather, they create increased exposure for the same problems that have traditionally plagued trademark holders in the era of human-driven brand development. Under Indian trademarks law, brand owners are entitled to sue a domain name registrant for infringement on a trademark for goods/services of the same class even if the former did not register the domain name. Section 29 of the Trademarks Act, 1999, imposes strict civil liability on any person who reproduces a registered trademark, including one chosen by a generative AI tool. Tools selecting brand names for trademarks in generative AI tools are subject to the same deceptive similarity tests as trademarks selected by humans, which means that the phonetic similarities in AI-assisted brand names carry increased risk in the context of trademarks with a large number of Indian buyers or a high level of visual similarity. The most logical response to the risk of collision is for brands to apply the same principles of trademark clearance to AI-assisted brand names as to names developed without AI assistance, including the same contractual allocation of responsibility for such clearance.
Pre-Launch Naming Protocol
Treat AI output as a longlist, not a shortlist – every AI-assisted name suggestion is treated the same as a hand-created suggestion in terms of undergoing a trademark search and other clearance procedures.
Conduct a full-class registry search – the name should be searched against the trademark registry in all relevant and possibly related classes for identical and similar trademarks, including phonetically or visually similar variations of the name, and pending applications.
Conduct a common-law search – the name should be searched against common-law trademarks, including checking trade directories, industry journals, and other sources for unregistered trademarks that may have common-law protection.
Cadila deceptive-similarity factors – apply the factors to any potentially infringing trademarks found in the above steps to determine if there is a likelihood of confusion, paying particular attention to the phonetic sound of the trademarks, and the class of purchasers and mode of purchase under Section 29 of the Trade Marks Act.
Multilingual search – check phonetic equivalents and meaning of the shortlisted names in Hindi and other Indian languages, where applicable.
Domain and social-media availability check – check availability of the name as a domain name and social-media handle, as a close equivalent in phonetics, and conduct the domain name check as a separate procedure from the trademark clearance process.
Structural-cluster clearance – where multiple variations on a root are generated, clear the entire cluster of names rather than just one variation.
Record responsibility for each step – ensure that the trademark clearance steps have been taken either by the client, the agency, or an outside counsel retained by either, before any names are submitted for trademark registration.
Trademark applications – apply for trademark registration for the shortlisted names.
Monitoring – register for watch services to monitor for registration of confusingly similar trademarks and domain names for at least the first 12–18 months after trademark filing.
Author: Gurjeet Singh Walia in case of any queries please contact/write back to us via email to content@khuranaandkhurana.com or at Khurana & Khurana, Advocates and IP Attorney
References
The Trade Marks Act, 1999 (India), ss. 27, 29, 103.
Cadila Health Care Ltd. v. Cadila Pharmaceuticals Ltd., (2001) 5 SCC 73.
Amritdhara Pharmacy v. Satya Deo Gupta, AIR 1963 SC 449.
The Information Technology Act, 2000 (India), s. 79.
Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Rules, 2021, as amended in 2026.
.IN Dispute Resolution Policy (INDRP), National Internet Exchange of India.
Uniform Domain-Name Dispute-Resolution Policy (UDRP), ICANN/WIPO Arbitration and Mediation Center.
United States Patent and Trademark Office, AI Strategy (January 2025).
Comments