When the Machine Borrows the Shelf : Liability for AI-Generated Trade Dress and Packaging Confusion in India
Introduction : Generative design software can generate scores of packaging ideas within minutes. A brand manager can ask for “a premium glucose-biscuit pack in warm yellow with child illustration” and get beautifully designed packages before lunchtime. However, these models are trained on extensive databases of images, including images of actual products placed on actual store shelves. Hence, an output generated by these models may unconsciously incorporate the very colour schemes, layouts and graphics associated with the leading product in the marketplace. When such a package makes its way to the shelves, the issue isn’t of whether the designer consciously copied, but whether the consumer is likely to be confused.
This blog answers four questions: which legal systems apply to the AI-generated packaging that looks like competitor’s trade dress; who is responsible; why a log of the prompts and audits of the model’s output will become crucial; and what kind of compliance process an organisation should follow prior to launching the product. The discussion will be conducted in the framework of Indian law, with the help of foreign precedents, where Indian precedent is lacking.
Legal Provisions
Trade Marks Act, 1999
“Mark” means a device, label, shape of goods, packaging and combination of colours, and a trademark must be capable of graphical representation and of distinguishing one trader’s goods from those of another. Infringement of a registered mark takes place if there is the use of an identical or similar mark in relation to identical or similar goods and there is likelihood of confusion, and under section 29(4), it extends to marks having a reputation in India in which use without due cause takes unfair advantage or is detrimental. Section 29(6) states that use means affixing the mark to goods or their packaging. Section 27(2) safeguards the common law right of passing off. Section 135 authorizes an injunction, damages or account of profits, but under section 135(3), such relief is confined to nominal damages if a defendant proves that it did not know of the mark belonging to the claimant at the time it first used it and stopped using it as soon as it learned of its existence. The offenses under sections 103 and 104 give the benefit of a defence of lack of fraudulent intention.
Copyright Act, 1957
Labels, logos and pack illustrations are artistic works as per section 2(c). In case of computer-generated works, the person causing the work to be created is considered as the author of the work under section 2(d)(vi). The reproduction of a substantial part of an artistic work without a licence constitutes an infringement as per sections 14(c) and 51.6
Designs Act, 2000
The design refers to the shape, configuration, pattern, ornament or composition of lines or colours applied to an article and regarded only as articles of eye; trademark and artistic work is not included in this definition. It is a piracy act as per section 22 to apply a registered design or an obvious imitation of that registered design on an article without consent. As per section 15(2) of the Copyright Act, the copyright expires in respect of a registrable but unregistered design after reproducing more than fifty articles by industrial process.
Unfair Competition, Consumer Law and Evidence
There is no Indian unfair competition law; however, Article 10bis of the Paris Convention requires member countries to control acts causing confusion about another’s products.1 The Consumer Protection Act, 2019 controls unfair trade practices and deceptive advertisement, which can be enforced by the Central Consumer Protection Authority. As regards evidence, Section 63 of the Bharatiya Sakshya Adhiniyam, 2023 regulates the admissibility of electronic evidence. Order XI of the Code of Civil Procedure, as amended for commercial disputes, mandates parties to disclose their relevant documents.
Legal Analysis
Liability Is Not Dependent on the Machine
These tests apply from the consumer's point of view, not the designer’s. Passing-off involves goodwill, misrepresentation, and damages, not fraud or fraudulent intent.11 Infringement asks whether there is likelihood of confusion from the perspective of an average purchaser with imperfect memory. As the business selling goods in AI-generated packaging is the user of the get-up in the context of trade, liability will not be avoided on account of the machine. The position of the provider is less clear. In Getty Images v Stability AI, the English High Court found that watermarks in the outputs from early versions of the model could be considered the provider’s own commercial message due to dependence on training data within its control, even though the infringement was described as historic and extremely limited. In India, an intermediary might claim protection from liability under section 79 of the Information Technology Act, 2000; however, it is unclear whether a system that creates its own output falls within the definition of “mere” intermediary. Liability is contractually assigned; almost all providers hold users liable for any output, while the liability protections offered to enterprise users come with certain conditions attached.
Passing off and Trade Dress
The Indian courts protect the get-up of packaging without registering the same. In Colgate Palmolive v. Anchor Health & Beauty Care, the Delhi High Court granted an injunction against the use of a colour combination of red and white on packages of toothpowder, finding that the trade dress, by itself, may constitute an indication of source.The Bombay High Court has protected distinctive bottle shapes using passing off action.The AI-related issue is the generalization tendency of models. Generative models generalize the conventions of the category, like green for mint or gold for premium products, and they are generally trade conventions. Liability will arise if the model output copies the distinctive combination of the get-up made by a particular brand. The other problem with such generous protection of get-up is that the leading brands in the category can get a monopoly over the trade convention of that particular category.
Copyright and Design
Copying is essential to the case of copyright infringement, whereas similarity is sufficient. Independent creation is a total defence. In the case of RG Anand v Delux Films, the Supreme Court of India questioned if the ordinary observer would have an unequivocal perception that the subsequent work was a copy.The issue becomes complex in the case of generative AI since there may be no direct causation between the infringement of copyright and the use of the design. Even though the user had no knowledge of the design by the rival party, the model could have been trained with it.
Uncertainty is also an issue regarding the rights of the business itself. Even though section 2(d)(vi) speaks about computer-generated works, Eastern Book Company v DB Modak states that a certain degree of creativity must be involved in the work based on the skill and judgment of humans,19 and US Court of Appeal, District of Columbia Circuit has said that copyright requires human authorship.A pack that is made using minimal involvement of a person may be hard to enforce through copyright, hence, registration of the pack either as a trade mark or design becomes more crucial. In the case of registered designs, the Full Bench of Delhi High Court in Carlsberg Breweries v Som Distilleries allowed a design infringement action along with passing off action to be taken in one suit.
The Importance of Prompt Records and Output Audits
Since liability is not contingent upon any form of intentionality, it may seem that the design process is entirely irrelevant. This is far from true. At the very least, the process record has relevance for five distinct matters.
To start with, it has relevance for honesty. If the prompt was one that asked a program to make a pack “like” one from a competitor, it is clear evidence of conscious copying, which the Supreme Court in Midas Hygiene v Sudhir Bhatia considered as sufficient reason to grant an injunction in an ordinary case of adoption.In turn, generic prompts and rejections of similar designs will indicate good faith. Secondly, it has relevance for remedies: section 135(3) requires evidence of the timing of awareness and subsequent cessation of use. Thirdly, it has relevance for criminal liability since acting in good faith and absence of fraudulent intent are a statutory defence here. Prompt histories are electronic records in Order XI, and their destruction after a controversy arises gives rise to an adverse inference under section 119 of the Bharatiya Sakshya Adhiniyam.
Moreover, the documents need to be admissible in court. As per Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal, the certificate under section 65B of the old Evidence Act was mandatory for any secondary electronic evidence, which requirement is reflected in section 63 of the Bharatiya Sakshya Adhiniyam. An appropriate audit would then involve recording information about the tool used and its version, all prompts, all generated responses, response selection and reasons thereof, any changes by humans, and the similarity check done.
Case Laws
Parle Products v JP & Co (1972) In this case, it was held by the Supreme Court that biscuit wrappers should be compared as a whole impression rather than side by side comparison in the eyes of a man of imperfect memory.
Cadila Health Care v Cadila Pharmaceuticals (2001) The Supreme Court provided a list of elements to determine deceptive similarity which includes the nature of marks, nature of goods and the class of purchasers.
Colgate Palmolive v Anchor Health and Beauty Care (2003) Trade dress is recognized by the Delhi High Court as being protected under passing off through trade dress, color combination etc.
Reckitt & Colman v Borden (1990) In this case, the get-up of the lemon shaped container was protected by the House of Lords and established the classical trinity of passing off.
Midas Hygiene Industries v Sudhir Bhatia (2004) The Supreme Court ruled that an injunction needs to be issued only after infringement or passing off, especially in cases of dishonest adoption.
Carlsberg Breweries v Som Distilleries (2018) In a five-member bench case of the Delhi High Court, the joining of design infringement and passing off in the same suit was allowed.
Getty Images v Stability AI (2025) The High Court of England ruled that there had been limited trade mark infringement by the generation of the claimant’s watermarks in the outputs.This is the first sign that generators might be equally responsible for infringement.
Implications for Practice
In the first place, the impact will be felt by fast-moving consumer goods, direct-to-consumer products and design agencies that have started using these generative technologies to reduce cost and time of development. The private label and challenger brands, which often try to steal ideas from the category itself, run the highest risks, because the difference between acceptable conventions and infringements is small. There is a need for proper warranties, indemnities and right of audit in contracts with agencies and vendors of AI technologies. In addition, there is a bigger policy debate, regarding the responsibilities of the provider of the output and the due diligence documentation by the courts, and India’s need for a specific look-alike packaging rule.
Pre-Launch Compliance Process
Quick Policy: use only sanctioned tools and prohibit prompts that mention competitor names, brand names and packaging, and create reference boards using only the company's own material.
Logging: keep logs of the tool, version of the model, its settings, all prompts used, choices and any human intervention with timestamps and hashes, keeping logs for at least the limitation period and during the product's entire lifecycle.
Trademark Clearance: Look up the trade marks database, including the device marks in Vienna classification and the designs database, and review the top products from the respective shelf.
Confusion Assessment: ask persons uninvolved in the project to evaluate the package design against competitor designs based on overall impression and imperfect recollection.
Legal Approval: get a written legal opinion for high-risk categories prior to printing.
Protection: apply for trade mark and design registrations for the package and document any human input in the creative process.
Response Post-Launch: observe the market and have a protocol for rapid investigation and immediate withdrawal upon receiving a complaint, maintaining the section 135(3) defence position.
Conclusion
Three conclusions emerge from this discussion. First, the use of AI is no different from any other technology in terms of the tests for trade mark infringement, passing off, copyright or design piracy; it is the business placing the pack into the market which will bear liability. Second, the process of designing, hidden in conventional litigation, is increasingly likely to come under closer scrutiny, as clear records and output audits become relevant to issues of dishonesty, remedies, criminal liability and authorship. Third, businesses using AI to create their packaging may have less rights in their designs than they think. Four reforms could help solve the problem. The DPIIT project to consider AI outputs must cover authorship and the causation requirement for design infringement in section 2(d)(vi). The courts and legislature need to clarify whether clearance records reduce remedies. Policy-makers need to examine transparency requirements for the providers of commercial generative design software and the possibility of similarity filters. Fourth, industry needs to devise a standard audit format for design records.
Author: Ajit Tada 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
Trade Marks Act 1999, s 2(1)(m), (zb).
Trade Marks Act 1999, s 29(1), (2), (4), (6)(a).
Trade Marks Act 1999, s 27(2).
Trade Marks Act 1999, s 135(1), (3).
Trade Marks Act 1999, ss 103, 104.
Copyright Act 1957, ss 2(c), 2(d)(vi), 14(c), 51.
Designs Act 2000, ss 2(d), 22; Copyright Act 1957, s 15(2).
Paris Convention for the Protection of Industrial Property (as amended 28 September 1979) 828 UNTS 305, art 10bis.
Consumer Protection Act 2019, ss 2(28), 2(47), 18.
Bharatiya Sakshya Adhiniyam 2023, s 63; Code of Civil Procedure 1908, Order XI, as amended by the Commercial Courts Act 2015.
Reckitt & Colman Products Ltd v Borden Inc [1990] 1 WLR 491 (HL).
Parle Products (P) Ltd v JP & Co (1972) 1 SCC 618; Cadila Health Care Ltd v Cadila Pharmaceuticals Ltd (2001) 5 SCC 73.
Getty Images (US) Inc v Stability AI Ltd [2025] EWHC 2863 (Ch); ‘Getty Images v Stability AI: What the High Court’s Decision Means for Rights-Holders and AI Developers’ (Mayer Brown, 20 November 2025) https://www.mayerbrown.com/en/insights/publications/2025/11/getty-images-v-stability-ai-what-the-high-courts-decision-means-for-rights-holders-and-ai-developers accessed 25 September 2026.
Information Technology Act 2000, s 79; Microsoft, ‘Microsoft Announces New Copilot Copyright Commitment for Customers’ (7 September 2023) https://blogs.microsoft.com/on-the-issues/2023/09/07/copilot-copyright-commitment-ai-legal-concerns/ accessed 25 September 2026.
Colgate Palmolive Co v Anchor Health and Beauty Care Pvt Ltd (2003) 108 DLT 51 (Del).
Gorbatschow Wodka KG v John Distilleries Ltd 2011 (47) PTC 100 (Bom).
RG Anand v Delux Films (1978) 4 SCC 118.
Department for Promotion of Industry and Internal Trade, Working Paper on Generative AI and Copyright: Part 1 (8 December 2025); ‘DPIIT’s Second Paper on Copyrighted Content in AI Likely in Two Months’ Business Standard (New Delhi, 11 December 2025) https://www.business-standard.com/technology/tech-news/dpiit-s-second-paper-on-copyrighted-content-in-ai-likely-in-two-months-125121100700_1.html accessed 25 September 2026.
Eastern Book Company v DB Modak (2008) 1 SCC 1.
Thaler v Perlmutter 130 F 4th 1039 (DC Cir 2025).
Carlsberg Breweries A/S v Som Distilleries and Breweries Ltd (2019) 256 DLT 1 (FB).
Midas Hygiene Industries (P) Ltd v Sudhir Bhatia (2004) 3 SCC 90.
Bharatiya Sakshya Adhiniyam 2023, s 119, illustration (g).
Arjun Panditrao Khotkar v Kailash Kushanrao Gorantyal (2020) 7 SCC 1; Bharatiya Sakshya Adhiniyam 2023, s 63.
Comments