THE EVOLVING ROLE OF ARTIFICIAL INTELLIGENCE IN INDIANINTELLECTUAL PROPERTY LAW: CHALLENGES, DOCTRINALGAPS, AND POLICY PATHS

Abstract

This paper examines the rising tensions between generative AI technologies and India’s human-centered IP regime in three key areas: copyright, patent and trademarks. It concludes that all three Acts – the Indian Copyright Act, the Patents Act, and the Trade Marks Act – make the assumption that humans write, invent and own things, which puts generative AI works in a regulatory no-man’s land. In copyright, human authorship is not recognized and Section 52’s limited list of fair-dealing exemptions adds to the confusion for text-and-data mining and AI-assisted creative works. A suggested practical principle is to require “significant human input” to find the limited protections bestowed to human input while allowing for AI assistance. In patent law, statutory definitions and office practices preclude AI from being a named inventor. Instead, “innovation oversight” – granting rights to those who have intellectual oversight over AI-generated inventions – could operate within the current laws. In trademark law, the requirements to own a trademark explicitly require the owner to be natural persons or legal persons. This means you must establish a contract and demonstrate that you are curating the marks yourself to protect and enforce your rights over the AI-Generated Marks. The essay also examines recent lawsuits (such as ANI Media Pvt. Ltd. v. OpenAI) and policy signals on licensing training data, which illustrate the importance of having clear TDM exclusions. Comparative examples from the UK, EU, and South Africa create possible trajectories for reform. Final recommendations are to clarify definitions of authorship and inventorship under the law, establish safe harbors for TDM, create best practices for documenting human contributions, licensing the AI training data, and establishing ownership agreements. These recommendations seek to realign India’s IP system with the rapidly changing landscape related to AI while also encouraging individuals to be inventive and create new things.

Introduction

Generative and assistive AI systems can now produce texts, images, logos, code, and technical devices that resemble human-generated items. For instance, Indian intellectual property laws that state the need for human creativity and a legal personality do not expressly cover non-human creators or inventors. This presents three immediate questions regarding: the “author” of autonomous outputs under the Copyright Act, the “inventor” under the Patents Act, and the owner of AI-generated marks under the Trade Marks Act when the revealers of the “creation” were a machine/s. Courts, offices, and policy articles all seem to agree: rights are the province of natural or juristic persons, and AI is only a tool. This means if a person wants to protect the outputs made-or-assisted by AI, they will have to demonstrate significant strategy and intellectual contribution to ownership and use contracts that clearly specify who owns what and what risks are created by ownership. Exploring India’s positioning on copyright, patents, and trademark and also signals on the legal and policy positioning on text-and-data mining (TDM) of AI training and provides doctrinal tests and compliance playbooks for academics and practitioners in the near future is crucial for this fast forwarding world.

Copyright: Human Authorship, AI Training, and the “Significant Human Input” Lens

Copyright law in India defines “author” in relation to persons and protects originality as a human intellectual creation. Recent commentary or litigation suggests that Indian law does not recognize non-human authorship, and therefore purely autonomous AI outputs (without sufficient human intellectual contribution) do not enjoy protection as “works.” Copyright can be claimed, however at least once a human demonstrates some creative control over the AI output, for example, by selection, arrangement, curation or modification.[1]

A prominent proceeding shaping the TDM debate is ANI Media Pvt. Ltd. v. OpenAI before the Delhi High Court. The court has entertained questions whether mass ingestion of copyrighted works for model training can be sheltered under Section 52 “fair dealing,” given India’s enumerated, relatively narrow exceptions.[2] Competing amici submissions reflect the divide: one view analogizes machine learning to “non-expressive” learning (more akin to human learning), while another partitions training into stages and treats data collection and training as expressive uses implicating reproduction rights under Sections 14 and 51.[3] No blanket safe harbour for training exists in India; DPIIT signals have emphasized authorization for training with copyrighted content, and practitioners widely advise licensing or relying on public domain and permissive sources for training corpora.[4]

With this in mind, a practical doctrinal framework is the “significant human input” concept: outputs that demonstrate significant human selection, editing, and authorial discretion may be appropriate for protection while outputs produced wholly autonomously would not. This perspective is consistent with the recent Indian academic and practice commentary and provides courts with some standard without change to statute – indeed, waiting on legislative modification.[5]

Key implications for stakeholders

• Creators and businesses should keep records of how people contributed to AI-assisted works and keep versioned edit histories to show that a person came up with the idea.
• AI developers who want to sell their products should make sure that the training data is properly licensed and that their own compliance evaluations are done on Section 52 exposure. The danger is bigger for commercial, large-scale TDM without permission.
• Policymakers who are thinking about a TDM exception may make it fit non-expressive applications and limited objectives, with the option to opt out and pay writers to keep their incentives. They could learn from the EU and UK debates while still following the stricter Indian fair dealing heritage.[6]

Patents: Inventorship, DABUS, and the “Innovation Oversight” Proposal

The Patents Act’s structure presumes a human inventor and limits applicants to “persons,” defined to include natural persons and certain juridical entities. In practice, the Indian Patent Office has objected to applications naming AI (e.g., DABUS) as inventor, citing Sections 2 and 6; the dominant comparative jurisprudence in the UK, US, and EPO also requires human inventorship. South Africa’s grant in the DABUS saga remains the outlier and is often explained by its formal filing system rather than a normative endorsement of AI inventorship.[7]

The doctrinal issue is not whether AI can technically help, but whether anything that doesn’t have legal personality or human intelligence can be a “inventor” and have the full set of exclusive rights that come with patents. Indian criticism and office practice align: AI cannot be designated as an inventor under existing legislation; rather, applicants must specify the actual and primary human inventor(s) or an assignee thereof. When AI significantly aids research and development, inventorship continues to identify those who originated the original thought and implemented it in a manner that meets the criteria for creative step and enablement.[8]

As a constructive path forward, recent scholarship proposes an “Innovation Oversight” approach: award patents to the individual(s) or entity exercising intellectual control over the overall inventive concept, assessing the degree of human oversight, problem framing, selection among AI-generated solutions, and integration into a patentable embodiment.[9] Where both “operator” and “innovator” contribute, ownership and revenue-sharing could be apportioned consistent with Section 50 and contractual arrangements. Crucially, this approach can be implemented with minimal statutory change by reading existing inventorship notions through the lens of human control and contribution.[10]

Compliance and drafting suggestions

• Maintain Keep up-to-date R&D documents that show how humans came up with the inventive step, how AI was used to come up with ideas or improve them, and the decisions made by humans that led to the claimed embodiments.
• Do not mention AI as the inventor; make sure there is a chain of title from the human inventors to the applicant through assignments that meet the requirements of Section 68.
• When AI is involved, expect problems with disclosure and sufficiency. Make sure the patent specification lets a POSITA do their job without relying on secret AI “black-box” magic, which means fulfilling enablement and best-mode requirements.

Trademarks: AI-Generated Logos, Distinctiveness, and Ownership

According to Section 2(1)(zb) of the Trade Marks Act, a mark must have the ability to be represented graphically and distinguish the goods/services of one person to those of another person. The Act employs the presumption of a proprietor that is either a natural or juristic person. AI doesn’t fit the categories of proprietor or author. So the application, generally, is likely made by the entity or person that was the client or user for which the logos, names or slogans were generated. Provided there is some contractual provisions and/or internal documents establishing ownership and control over the generated output. If there is no statutes granting specific recognition for AI creators, then owneship for marks generated would hinge on whatever humans selected, modified or used as an indicator of source ownership.[11]    

Practical cHALLENGES and strategies

• UNIQUE

Many AI methods create stock or convergent messages, which increases the likelihood of Section 9 challenges. To improve the distinctiveness of your brand, consider human editorial discretion, brand style manuals, and proof of use. If the inherent distinctiveness is low, use evidence regarding marketing, sales, and/or recognition to develop a file for acquired distinctiveness.

• CLEARANCE

AI can copy existing marks, conduct pre-filing, comprehensive searches, and avoid ‘prompt collisions’ with other registered or famous marks. AI can also analyze internal prompts and outputs to forecast likelihood of confusion and the degree to which a mark might be descriptive.

• OWNERSHIP

Make sure the business—not the tool vendor—owns what AI generates, align platform user agreements, contractor agreements, and corporate IP policies. Make sure assignments or licenses are secure, and that your vendors or platforms cannot claim ownership to what you have generated.

• ENFORCEMENT

Establish a record of human curation and brand adoption in case you face threshold issues in oppositions, invalidity challenges, or enforcement actions based on lack of distinctiveness or authorship.[12]

Training Data, Fair Dealing, and Regulatory Signals

Indian law does not explicitly provide a TDM exception for AI training. The Section 52 exceptions that are spelled out usually receive a narrow interpretation. Commentary suggests that large-scale scraping and storing of materials with the commercial purpose of training models is likely to implicate reproduction rights under Sections 14 and 51 unless a license is obtained. Some argue that forms of training which take place as tokenization/vectorization would be “non-expressive” and therefore less likely to infringe any reproduction rights. The Delhi High Court’s consideration in ANI v. OpenAI[13] and policy comments attributed to the Department of Promotion of Industry and Internal Trade point to a developing trend towards requiring authorization for training with protected content, placing India closer to EU style permission structures and farther from the broader US approach of fair use.

COMPLIANCE PLAYBOOK FOR DEVELOPERS AND DEPLOYERS

• Prefer datasets subject to licensing, public domain works, and CC/permissive works intended for use conforming to a scope of training use. Maintain a registry of rights and data lineage.
• Query the nature of data use within a given phase and cohere data-use categories in separate phases. Characterize the data collection phase or retention of data for use in training separately from the added latency of training the output to conform to user queries. Categorize collection, storage, and training as a potential high-risk activity in relation to infringement and while developing a retention system consider enforcing data minimums and honoring user opt-out in use cases where possible.
• Implement output calibration to reduce expressive regurgitation and substitution effects that might give rise to claims of infringement or passing-off.
• For enterprise deployments, contract for warranties, indemnities, and audit rights regarding training data provenance and claims arising from third parties.[14]

Comparative Notes and Indian Trajectory

Comparative touchpoints matter for Indian courts and policymakers. The UK Supreme Court has confirmed no patent inventorship for AI (DABUS), and EU’s copyright directive has specific TDM provisions. Indian office practice and court posture have so far mirrored the human-centric baseline in patents and a cautious stance on copyright exceptions.[15]

Academic and practice literature in India increasingly recommends

• A statutory TDM exception with scope, purpose limits, remuneration, and opt-out mechanisms.
• A statutory acknowledgment of AI-assisted works coupled with a “significant human input” test for authorship.
• Patent guidance clarifying the inventorship whereas AI assists the conception and reduction to practice for operationalizing an “innovation oversight” test through rules or manual updates rather than immediate legislative overhaul.[16]

SUGGESTIONS

Research focus areas for a law review-style article

• Doctrinal fit

To determine which of the current definitions of “author,” “creator,” or “proprietor,” if any such current definitions are adequate to rely on tests developed in the courts, or whether amendments to statutes by parliament are warranted. For example, a rights-holder can defend proprietary rights in copyright ownership through a significant human input standard, compared to “innovation oversight” in the patent process. Additionally, administering the significant human input standard could vary across various fact scenarios.

• REMEDIES AND LIABILITY

Consider the issue of secondary liability for platform providers where acts of infringement occur because of AI output, and risk allocation through licensing and indemnity, and/or the platform terms of service.

• EVIDENCE LAW

Develop evidentiary procedures for achieving the standards of evidence required to demonstrate, for IP litigation, a human contribution, a dataset licensing, and provenance (evidenced as persons contributing). Types of evidentiary standards might include version control logs, prompt-output logs, editing histories, and data registries.

• COMPETITION LAW AND ACCESS

Consider the over-enclosure risk where information resources require licensing of a potentially massive corpus of data provided to train Ai or under the contract theory (to the risk of the creator markets); for TDM access, consider unanimous licensing or extended collective licensing in India.[17]

DRAFTING AND COMPLIANCE CHECKLISTS FOR APPENDICES

• Copyright

(a) Human contribution documentation and

(b) Output after review for originality; training data license matrix; Section 52 risk matrix

• Patents:

(a) Inventorship worksheet;

(b) Ownership chain-of-title documents (as per Section 68)

(c) Enablement and sufficiency checklist for AI-assisted which apply

(d) Invention disclosure templates for capturing human conception.

• Trademarks:

(a) Clearance search logs;

(b) Steps undertaken to enhance the distinctiveness and

(c) Net platform Terms Of Service (TOS),

(d) An evidence plan for acquiring distinctiveness[18]

Conclusion

Intellectual property law in India is mostly based on human agency. That basis should remain consistent—but it can be reasonably expanded to include AI through administrable tests that maintain incentives and decrease uncertainty. A judicial approach before the end of the year that applies a “significant human input” test in copyright and an “innovation oversight” test in patents (alongside contractual allocation and provenance practices) can stabilize rights while Parliament considers TDM-focused legislation and additional targeted definitional clarifications. Proprietorship for trademarks must stay with human or juristic actors, while distinctiveness and clearance can be raised to take on converging AI outputs. Until statutory reform comes, rigorous documentation, licensing hygiene, and conservative deployment will be required for differentiation between the output of protectable innovation in India’s AI period.

Reference(S):

[1] Generative AI and Copyright Issues, drishtiias, (July 17, 2025, 10:30 PM), https://www.drishtiias.com/daily-updates/daily-news-analysis/generative-ai-and-copyright-issues; Maheshwari & Co., AI Generated Trademarks in India: Legal Issues, MAHESHWARI & CO. (July 15, 2025, 10:30 PM), https://www.maheshwariandco.com/blog/ai-generated-trademarks-in-india/.

[2] Pragya Jha, Bernd Justin Jütte (University College Dublin), Does Human Learning equal Machine Learning? High Court of Delhi to rule on lawfulness of TDM for Machine Learning, (Oct 7, 2025, 10:22 AM), https://legalblogs.wolterskluwer.com/copyright-blog/does-human-learning-equal-machine-learning-high-court-of-delhi-to-rule-on-lawfulness-of-tdm-for-machine-learning/.

[3] Pallavi Rao, Soumya Dasgupta & Siddharth Kothari, ‘Fair Use’ in the Age of AI, (Oct 7, 2025, 10:27 AM), https://corporate.cyrilamarchandblogs.com/2025/04/fair-use-in-the-age-of-ai/.

[4] Copyright Implications in training Artificial Intelligence (AI) Models, Mohit Porwal, Associate Partner

Krupa Vyas, Associate, (Oct. 7, 2025, 10:22 AM), https://chambers.com/articles/copyright-implications-in-training-artificial-intelligence-ai-models.

[5] Nikhil Mishra1and Digvijay Singh , AI-Generated Work and its Implications on Copyright Law in India, (Oct 7, 2025, 10:27 AM), Journal of Intellectual Property Rights Vol 30 January 2025, pp 35-44 DOI: 10.56042/jipr.v30i1.5862, file:///C:/Users/hp/Downloads/JIPR-209+corrected+proof.pdf.

[6] Chambers and Partners, Copyright Implications in Training Artificial Intelligence (AI) Models, (Oct. 7, 2025,10:44 AM), https://chambers.com/articles/copyright-implications-in-training-artificial-intelligence-ai-models.chambers; Mondaq, Generative AI And Trademarks: The Need For Legislative Intervention (July 20, 2025, 10:44 AM), https://www.mondaq.com/india/trademark/1653332/generative-ai-and-trademarks-the-need-for-legislative-intervention.mondaq.

[7] Khurana & Khurana, DABUS Case: AI Inventorship in Indian Legal Regime, KHURANA & KHURANA BLOG, (Oct. 7, 2025, 10:47 AM), https://www.khuranaandkhurana.com/2025/03/19/dabus-case-ai-inventorship-in-indian-legal-regime/.khuranaandkhurana.

[8] Surana & Surana, Legal Crossroads: The DABUS Paradox in Patent Law, SURANA & SURANA, (Oct. 7, 2025, 10:47 AM), https://suranaandsurana.com/legal-crossroads-the-dabus-paradox-in-patent-law/.suranaandsurana

[9] Supra note 1, Drishti IAS

[10] Indian Journal of Law & Technology, Balancing Indian Copyright Law with AI-Generated Content: The Significant Human Input Approach, IJLT BLOG, (Oct. 7, 2025, 10:30 PM), https://www.ijlt.in/post/balancing-indian-copyright-law-with-ai-generated-content-the-significant-human-input-approach.   

[11] Daisy Banakhede, AI-Generated Trademarks: Legal Ownership and Challenges in India, (May 10, 2025, 10:30 PM),  https://www.iiprd.com/ai-generated-trademarks-legal-ownership-and-challenges-in-india/

[12] Akansha Singh, Urvi du, Generative AI And Trademarks: The Need For Legislative Intervention, MONDAQ,  (July 5, 2025. 10:30 PM), https://www.mondaq.com/india/trademark/1653332/generative-ai-and-trademarks-the-need-for-legislative-intervention.

[13] Ranjan Narula, Parth Bajaj, ANI v OpenAI: generative AI’s use of copyrighted works under Indian law,

(July 22, 2025, 10:30 PM), https://www.managingip.com/article/2f3huy6goqztxnabhdvy8/sponsored-content/ani-v-openai-generative-ais-use-of-copyrighted-works-under-indian-law.

[14] Harshal Chhabra and Arihant Sethia, Forum NLSIU, The Impact of Artificial Intelligence on Intellectual Property Rights: A Case for Reform in Indian Patent Law by “Innovative Oversight” Approach, NLSIU IJLT BLOG (Apr. 10, 2025, 10:30 PM), https://forum.nls.ac.in/ijlt-blog-post/the-impact-of-artificial-intelligence-on-intellectual-property-rights-a-case-for-reform-in-indian-patent-law-by-innovative-oversight-approach/.  

[15] Vedika Chawla, UK Supreme Court Confirms No Patent for “AI-invented” Inventions, SpicyIP, (Dec. 22, 2023, 10:30 PM), https://spicyip.com/2023/12/uk-supreme-court-confirms-no-patent-for-ai-inventions.html.

[16] Akshat Agrawal, Indian Copyright Law and Generative AI: Part 2- Transformative and Extractive Use, IPRMENTLAW, (May 29, 2024, 10:30 PM), https://iprmentlaw.com/2024/05/29/indian-copyright-law-and-generative-ai-part-2-transformative-and-extractive-use/.

[17] Mohit Porwal, Associate Partner and Krupa Vyas, Copyright Implications in training Artificial Intelligence (AI) Models, CHAMBERS AND PARTNERS, (Apr. 14, 2025, 10:30 PM), https://chambers.com/articles/copyright-implications-in-training-artificial-intelligence-ai-models.

[18] Maheshwari & Co., AI Generated Trademarks in India: Legal Issues, MAHESHWARI & CO. (July 15, 2025, 10:30 PM), https://www.maheshwariandco.com/blog/ai-generated-trademarks-in-india/.