Kopal Agarwal
Student
Faculty of Law, University of Delhi
Email: kopalag27@615gauravyadav
Abstract
Trademarks laws have over a hundred years depended upon an apparently straightforward concept that a mark is a badge of origin recognised by the eye and memory of the average imperfectly recollecting consumer. The Artificial Intelligence system challenges this assumption at its core. This paper analyses the consequent clash between AI and Trademark laws, mainly from three perspectives – the efficiency improvement AI introduces into searching and clearance, the doctrinal disturbance caused by the same in average consumer and the sufficiency of current statutory framework. Drawing on recent Indian and comparative jurisprudence on deepfakes, keyword advertising and algorithmic impersonation, this article contends that Indian trademark law needs to develop an innovative technology sensitive confusion test and proprietorship criteria for machine made marks, while never losing sight of its human-centric approach to trademark law.
Keywords: Artificial Intelligence, Trademark Law, Deceptive Similarity, Algorithmic Confusion, Average Consumer, Keyword Advertising, Honest Concurrent Use, Cybersquatting.
Introduction
With the dawn of the fourth revolution, AI and man started being inextricably linked, marking the final completion of materialization of replicating human intelligence on machines. The transformation has been aptly put ahead by venture capitalist Marc Andreessen when he said, “AI is new electricity.” This transformation reaches down to the field of trademark law where AI’s abilities are moulding the basic traditional practices and legal principles.
Trademarks are an easier way of indicating the quality and characteristics of a product, saving the consumers ample time that would be spent in research. The trademark acts as an indicator of origin for specific goods and services. A trademark goes far beyond a simple logo as it could consist of a letter, a number, a word, a sentence, a sound, a smell, a shape, a symbol, packaging, a movement or even any combination of these elements. Trademark law is designed to serve as protection against anyone attempting to counterfeit products marked with a unique identifier and then pass them off as real products made by specific manufacturers But the AI at times overcomes the trademark law.
The Development of AI in Trademark Administration
Machine learning and Natural language processing models have made it possible for AI to revolutionise the trademark management process in terms of increasing its speed and accuracy.
Trademark laws can trace their origin to the time when consumer behaviour was characterized by the role played by shop assistants, who acted as intermediaries between buyers and unbranded commodities, giving them recommendations according to the latter’s quality. Later on, supermarkets emerged where consumers began to decide about purchases on their own, taking into consideration the well known name of a product. Much later, with the advent of social media, there appeared influencers, who also played the part of intermediaries by trying out products and proving the correctness or incorrectness of sellers’ claims.
Thus, trademark legislation provided the goal of protecting the ‘average consumer’ from misleading regarding the product that looked like his or her favourite brand but actually was its cheap imitation and also of protecting the seller from having his or her identity stolen by third parties producing counterfeit goods. However, AI system is almost a ‘perfect consumer’ because it can pinpoint the real source of the product with absolute accuracy without creating any confusion of brands and thus diluting the role played by trademark law.
The Silver Lining: Positive Impacts of AI on Trademark Governance
- Flagging deceptive similarity risks
Where a manual comparison of a proposed mark against a national register of several million entries was once the work of days, algorithmic similarity engines can generate a ranked list of deceptively similar marks almost instantaneously, allowing brand owners and registries alike to identify potential conflicts far earlier in the adoption cycle. This preemptive risk flagging function reduces both the volume of avoidable disputes reaching the Registry and the incidence of costly rebranding after launch. - Detecting counterfeiting listings on e-commerce platformsPlatforms such as Amazon, Flipkart and Alibaba use AI-driven image recognition, listing-text analysis and seller-behaviour modelling to detect counterfeit or infringing listings at a scale impossible for human moderators, taking down millions of suspect listings before a single sale occurs. These systems compare product images, packaging and seller metadata against verified brand registries, allowing platforms to intercept infringement at the point of listing rather than after consumer harm has occurred.
- Controlling Cybersquatting
AI powered domain-monitoring tools scan newly registered domains, alerting brand owners to potential cybersquatting long before a complaint would otherwise be triggered. This continuous and automated surveillance has meaningfully shortened the window in which a bad faith registrant can exploit a confusingly similar domain name.
The Shadow Side: Negative Implications of AI for Trademark Law
- False Narratives
The same systems that accelerate clearance searches are also capable of confidently generating incorrect results. These include false positives that prevent the adoption of a mark unnecessarily and false negatives that allow adoption when a human searcher would reject it. - The Doctrine of Average v/s Perfect Consumer
There are certain doctrines which are considered the bedrock of trademark law. These include doctrine of confusion, initial interest confusion, post purchase confusion, and the doctrine of the average consumer with poor recall abilities. Not only are these doctrines helpful in avoiding fraud but they also form the base on which market based doctrines such as discoverability, brand loyalty, after sales service and brand reputation can prosper. By doing so, trademark law allows for an emotional bond to develop between the buyer and the seller.
Average consumer forms the foundation of trademark law and they need to be revisited in the light of these rapid advancements in technology. The Supreme Court of India explained in the case of Cadilla Healthcare Ltd. v. Cadilla Pharmaceuticals Ltd. that a consumer can be termed as an average consumer if he has average intelligence and the tendency of imperfect recollection. This explanation is in utter contradiction with the essence of AI application. An AI based bot does not possess average intelligence, nor is it capable of imperfect recollection. AI driven retail and advertising systems behave like a perfect consumer as they compare marks pixel-for-pixel, index every keyword, and never forget a prior search - Keyword Advertising
Keyword advertising works using a straightforward process whereby sponsored listings are provided when the search input contains a keyword that has been purchased by a business. With each click on the sponsored link, the party intermediating between the ad and the user makes money. Competitors can, therefore, use a brand’s goodwill against it to lure customers who were trying to access one brand but are directed to a different brand instead.
The case of Lush v. Amazon is a classic example of a court’s ruling on such a matter. Amazon was sued by Lush for having bought the keyword ‘Lush’ using Google’s bidding service. Thus, when users searched for ‘Lush’ on Google, their traffic was redirected to Amazon. On Amazon’s platform, however, its recommendation algorithm showed other products instead of Lush products even though the search had been done based on the Lush trademark. Thus, Amazon was held liable for trademark infringement.
In Indian courts, the position adopted regarding similar issues is one that is more nuanced and evolves over time:
Consim Info Pvt Ltd v. Google India Pvt Ltd: In this case, the trademarks that were at stake included generic or descriptive words like language, community and matrimony. Taking into consideration the above mentioned circumstances, the Court was reluctant to impose an injunction since it would disrupt the overall scheme of things for Google, thereby creating a monopoly in the matter of search results for the appellant. There was also no indication of confusion.
Hindware Ltd. v. Grohe India Pvt. Ltd. & Ors.: In this particular case, the core issue was that the famous trademark ‘Hindware’ had been used illegally by its competitors in the Google AdWords system. As such, when someone searched for anything related to Hindware, advertisements of competitors came up using that keyword. Consequently, the court held that the way Google used the trademark in its commercial capacity amounted to trademark infringement. Therefore, Google could not use the trademark or even any variation thereof as a searchable keyword.
- Algorithmic Confusion
There is also another form of distortion that can take place, which occurs not through mediation but from the actions of the AI itself. Through autocomplete and “customers also bought” suggestion engines, the competitor’s brand will often be suggested in reply to the search for the brand of the trademark holder. There is also the issue of voice assistance, wherein phonetic similarities between brand names, especially in terms of different accents or languages, can lead to a situation where the assistant orders, recommends, or reads aloud the competing product at the request of the trademark holder. - Brand Impersonation
Deepflake videos and audios using Generative AI may be used to create fake endorsements and advertisements from celebrities and companies which will bear the company’s logo and trade dress, whereas AI website generators can replicate the colour scheme and trade dress of competitors at a fidelity and speed that manual copying never permitted. - AI Generated Logos
Generative design tools can now produce an original-seeming logo from a short text prompt in seconds. This convenience conceals a genuine doctrinal puzzle regarding who will be the proprietor of that logo. Furthermore, it also violates Section 12 of Indian Trademarks Act, 1999 which talks about honest and concurrent use of the mark.
The Statutory Liability for AI Generated Trademarks
Section 27 and Section 29 of The Indian Trademarks Act, 1999 provide for passing-off and infringement of the mark, respectively. However, The question of liability becomes complicated when AI generated content infringes on trademarks. The parties facing liability may include:
AI Developers: In some cases, developers of AI models might be found liable for any infringement by their models in situations where they knew or could have predicted that there was such a possibility but did not take necessary precautions. Many AI developers have already faced trademark infringement lawsuits where their models reproduced trademarks from the training data sets they used.
Users: People who use AI based software for the purpose of creating content are not immune from liabilities either. Persons or entities who intentionally or negligently create any infringements by use of the software may be liable for their actions. Liability will largely depend on two elements: the extent of the control exerted by the user over the output of the AI software and knowledge of the potential infringement.
Platforms: There is also the question of whether the platform upon which AI-generated content was posted can be exposed to liability, particularly if the platform had knowledge of the infringement and failed to take action. In such a situation, traditional legal theories such as contributory liability and vicarious liability will apply.
International Framework
There is no existing international treaty covering AI-generated trademarks, as well as an inconsistent global reaction in the form of soft law guidance and platform self-regulation.
Europe: EUIPO has incorporated AI-based visual and sound searching into its trademark view and similarity search tools, while the EU Digital Services Act places the duty of care for counterfeits detected by AI upon online marketplaces and the EU Artificial Intelligence Act, which although does not specify trademarks per se, places obligations of transparency on high-risk and generative AI systems, which will include the labelling of AI-generated marketing and brand materials in the future.
United States: USTPO has developed internal guidelines on AI assistance in the examination process and has started using machine learning algorithms to detect possible conflicts during the search procedure.
China: The most aggressive platform-level response comes from China, with its dominant players Alibaba having rolled out AI systems for detecting counterfeits as a condition of access to their markets, thus representing the enforcement response under government directive as opposed to case law-based doctrinal developments.
At the multilateral level, the WIPO maintains the Global Brand Database as an AI enabled search tool and has convened the ongoing WIPO Conversation on Intellectual Property and Artificial Intelligence, but this remains a policy dialogue rather than a binding instrument. The overall picture is one of convergent practical adoption, leaving India free, and arguably obliged, to develop its own coherent domestic answer.
Policy Recommendations: Balancing AI Innovation and Trademark Protection
- An Ownership Test for AI Created Marks: The Act must be modified or guided by the government in such a way that ownership of any AI created mark lies with the natural person or legal person who directs the generative process and first adopts the output in trade.
- A Hybrid Test for Digital Marketplace Confusion Involving Algorithms: The Courts and Registry must be made to recognize, besides the existing test based on average consumer confusion, a hybrid algorithmic mediation test for keyword advertising, autocomplete suggestions, and voice-assistant responses on not merely what an average consumer will notice, but also what confusion is likely to arise from the algorithms used by the digital marketplace platform itself.
- Conditional Safe Harbour for Platforms: Safe harbour for intermediaries as provided by the Information Technology Act needs to be made conditional for larger e-commerce and advertising platforms, where there are AI-based systems that can detect counterfeit and deepfake content, along with rapid takedown policies.
- Algorithmic Audit and Complaints Procedure at the Registry: There needs to be an internal mechanism in the Trade Marks Registry to periodically check the AI augmented search and examination mechanisms for false positives and false negatives and a clear complaints mechanism for those who believe there is a mistake in the algorithmic recommendation.
Conclusion
The collision between Artificial Intelligence and trademark law is not a distant possibility but a lived reality. The integration of AI into the process of trademark protection has made it more efficient, fast-acting and vigilant in the face of counterfeiting threats but it has also introduced a relentless intermediary, perfectly focused on every aspect of trademark, between the trademark and its intended consumer, as well as starting to develop new trademarks itself, thus shaking the connection between the creation, use and ownership of trademark which until now had been the hallmark of this field. The silent handshake between producer and consumer must still be honoured, mainly when a machine is the one extending the hand.
References
- Elizabeth Rocha, “Sophia: Exploring the Ways AI May Change Intellectual Property Protections ,” DePaul Journal of Art, Technology and Intellectual Property Law https://via.library.depaul.edu/cgi/viewcontent.cgi?params=/context/jatip/article/1588/&path_info=28_2_3_Sophia_Exploring_the_Ways_AI_May_Change_Intellectual_Property_Protections.pdf
- How Predictive AI will change shopping https://hbr.org/2016/11/how-predictive-ai-will-change-shopping
- Cadwallader, Clint D, “AI Unleashed: Navigating Intellectual Property Rights in the Age of Artificial Intelligence” Nebraska Lawyer, Vol. 27, Issue 3 (2024), pp. 5-14 https://www.mcgrathnorth.com/16D6CA/assets/files/documents/TheNebraskaLawyer-May-June2024.pdf
- Cadilla Healthcare Ltd. v. Cadilla Pharmaceuticals Ltd., (2001) 5 SCC 73
- Ishan Patel, Exploring the Impact of Artificial Intelligence on Trade Mark and Copyright: Challenges and Opportunities https://www.ijrti.org/papers/IJRTI2505200.pdf
- Cosmetic Warriors Limited and Lush Ltd. v. Amazon. Co. Uk Limited and Amazon Eu Sar
- Idan Mosseri, Trademarker – Artificial Intelligence based similarity search engine https://www.researchgate.net/publication/333652029_TradeMarker_-_Artificial_Intelligence_based_Trademarks_Similarity_Search_Engine
- Consim Info Pvt. Ltd. v. Google India Pvt. Ltd. (2010) 6 CTC 813
- Hindware Ltd. v. Grohe India Pvt. Ltd. & Ors. CS(COMM) 591/2017
- WIPO Artificial Intelligence based search engine tool https://www.wipo.int/pressroom/en/articles/2019/article_0005.html
