A technology company headquartered in Singapore deploys a generative AI platform to produce marketing content, software code, and product imagery for its Indian subsidiary. The output is commercially valuable. The company assumes it owns all intellectual property in those outputs. Then a competitor reproduces a substantial portion of the AI-generated material. Additionally. The company's legal team discovers that no clear answer exists under Indian law as to who. if anyone. actually holds enforceable rights over what the machine created. The financial loss is real. The legal position is unresolved.
AI-generated works in India occupy a doctrinal gap. Indian intellectual property legislation requires a human author or inventor, but does not expressly address outputs produced autonomously by machine learning systems. Courts have not yet delivered definitive guidance. Businesses deploying AI in India must structure their workflows, contracts, and filings carefully to secure the strongest available protection under the current body of law.
This analysis examines the doctrinal foundations of that gap, the competing interpretations emerging in practice, the distance between statute and commercial reality. The cross-border dimensions relevant to international businesses. Additionally, the strategic steps available to companies operating across the Asia-Pacific and Middle East region. It also considers the regulatory trajectory and what practitioners and in-house counsel should monitor over the next twelve to twenty-four months.
Doctrinal foundations: what the existing body of law actually says
Indian intellectual property legislation developed in an era when authorship and inventorship were self-evidently human concepts. The copyright legislation defines "author" by reference to categories of natural persons – the person who creates a literary work, the composer of a musical work, the photographer. The patent legislation requires that an inventor be identified, and the office consistently treats that requirement as demanding a human individual.
Neither statute was drafted with generative AI in mind. The result is a body of law that answers the wrong question. It tells us who among a group of human collaborators holds rights. It does not tell us what happens when the creative or inventive act is performed substantially or entirely by a machine operating on learned parameters rather than conscious direction.
The closest available doctrinal anchor is the concept of the "work made for hire" and its civilian analogue in employer-employee relationships. Where a human employee uses an AI tool as an instrument – directing it, selecting outputs, applying editorial judgment – Indian copyright doctrine permits treating the employer or commissioning party as the rights holder. This reasoning draws on the same logic that attributes authorship of a photograph to the photographer rather than to the camera.
The instrument theory has intuitive appeal. It breaks down, however, at the threshold of autonomy. When an AI system generates a novel, drafts a patent claim. Alternatively, composes a melody with minimal human direction beyond the initial prompt. Calling that human a "creator" in any meaningful sense requires considerable conceptual stretching. Courts in India have not yet confronted a case that forces resolution of this boundary. The absence of judicial authority leaves businesses – and their technology licensing arrangements – exposed to a range of possible outcomes.
Patent doctrine presents a parallel difficulty. The patent legislation centres on inventive step and the non-obviousness requirement. Indian patent examiners apply these requirements against the standard of a person skilled in the art. An AI system that identifies a non-obvious chemical compound or engineering configuration has no "skill in the art" in the statutory sense. More fundamentally, the human inventor requirement means that a company cannot simply name its AI system as inventor. It must identify the human or humans whose contribution crosses the threshold of inventive activity.
In practice, this leads to a filing strategy that emphasises the human engineers and data scientists who designed the AI model, curated training data, and defined the problem space. Whether that contribution satisfies the inventive step requirement is a separate and genuinely contested question. Examiners in India have raised objections in AI-adjacent cases on grounds that the claimed contribution lies in the model's outputs rather than in the human applicant's intellectual act. Practitioners note that these objections are increasingly common and that response strategies require careful framing of the human contribution at each stage of the development process.
Competing interpretations and the gap between statute and practice
Three interpretive positions compete in the current Indian legal environment, and none commands consensus.
The first position holds that AI-generated outputs receive no intellectual property protection unless a human author or inventor can be identified who made a genuine creative or inventive contribution. Under this view, an AI-generated novel or a machine-identified compound falls into the public domain from the moment of creation. Rights holders who rely on copyright or patent protection for their AI outputs are, on this analysis, holding a paper right that a court might decline to enforce.
The second position applies a purposive construction of existing legislation. It argues that intellectual property legislation should be read to achieve its underlying objective. incentivising investment in creative and inventive activity. rather than to exclude entire categories of output simply because human authorship is attenuated. On this reading, the entity that invests substantially in AI infrastructure, training, and deployment should be treated as the constructive author or assignee of the resulting outputs. This position finds some support in the general principles of Indian intellectual property law relating to employer-created works and commissioned works.
The third position, which represents the practice of a significant share of well-advised companies, treats the question as manageable through contract and commercial structuring rather than through statutory interpretation. Technology licensing agreements, employment contracts, and service agreements are drafted to assign any rights that may exist – whatever their ultimate legal characterisation – to the desired party. Trade secret protection is layered over copyright and patent claims. Confidentiality obligations prevent competitors from accessing the outputs long enough for the company to establish commercial advantage.
The gap between statute and practice is most visible in enforcement. A company that has built its IP portfolio around AI-generated outputs may find, when it seeks to enforce rights against a copyist, that the defendant raises a "no valid copyright" defence. The burden then shifts to the plaintiff to demonstrate sufficient human authorship. Courts examining that question will have no clear precedent to guide them. The risk is not merely theoretical: it materially affects how technology licensing fees are priced, how investors value IP-heavy AI companies, and how acquisition due diligence must be conducted.
Algorithmic accountability adds a further layer of complexity. When an AI system produces content that infringes a third party's copyright – because it reproduces substantial portions of training data – the question of software liability arises. Who is responsible: the developer of the model, the company that deployed it, or the user who submitted the prompt? Indian tort law and contract law provide partial answers, but the interaction with intellectual property legislation has not been tested in contested litigation. Businesses operating AI-as-a-service models in India carry exposure that their indemnity clauses and technology licensing terms may not fully address.
For a comparative perspective on how similar questions are being resolved in a neighbouring high-growth market. See our analysis of AI-generated works and intellectual property in the UAE. There, a distinct regulatory model is emerging under the DIFC and federal digital economy legislation.
Training data, fair dealing, and the infringement exposure businesses overlook
A question that receives less attention than authorship – but carries at least equal commercial risk – is the status of training data under Indian intellectual property legislation. Large language models and generative AI systems are trained on vast corpora of text, images, code, and other content. Much of that content is protected by copyright. The question of whether training an AI system on copyrighted material constitutes infringement is unresolved in India.
Indian copyright legislation contains a fair-dealing provision. It permits use of protected works for purposes including research and private study. Whether large-scale commercial AI training qualifies as research in this sense is genuinely contested. The fair-dealing doctrine in India is narrower than the fair-use doctrine in the United States. It applies to specific permitted acts rather than to a broad balancing test. Practitioners and courts in India have not yet determined whether the commercial purpose of AI training disqualifies a training activity from fair-dealing protection.
The practical consequence is that a company that trained its AI model on internet-scraped content. or that deployed a third-party model trained on such content. may face infringement claims from rights holders whose works appeared in the training corpus. Those claims would be pursued through civil litigation before the High Courts. Digital services companies and AI platform operators should assess their exposure before infringement proceedings are commenced against them.
A non-obvious risk arises at the output stage. If a generative model produces content that closely resembles a training data item. a passage of text, a visual style. A melody. the output itself may constitute infringement, regardless of whether the training stage was lawful. Indian courts apply a substantial similarity test. That test does not require identical reproduction. A court could find infringement even where the AI system produced the output without any explicit instruction to reproduce the original.
Businesses deploying AI systems in India should conduct a training data audit. They should identify which model or models underpin their products, assess the publicly available information about those models' training corpora, and obtain contractual representations from model suppliers about the steps taken to manage copyright exposure. Technology licensing agreements with model providers should address indemnity obligations specifically in relation to training data infringement claims. Many standard AI platform agreements currently do not.
To explore how intellectual property protection can be structured effectively for AI-assisted businesses operating in India, our dedicated advisory practice covers the full spectrum of available instruments: intellectual property services in India.
Cross-border implications for Asia-Pacific and Middle East businesses
International businesses operating between India and other Asia-Pacific or Middle Eastern jurisdictions encounter a multi-layered problem. The intellectual property position in the jurisdiction of creation, the jurisdiction of deployment, and the jurisdiction of enforcement may all differ. A work created by an AI system in India and distributed in Singapore, Japan, or the UAE may attract different levels of protection in each market.
India is a signatory to the principal international intellectual property conventions. These conventions operate on the principle of national treatment: a rights holder is entitled to the same protection in each member country that the country extends to its own nationals. However, national treatment does not resolve the underlying question of whether a right exists in the first place. If Indian law does not recognise copyright in an AI-generated work, there is no right to be extended under national treatment to a foreign plaintiff.
Companies structuring cross-border AI deployments should therefore identify the jurisdiction whose law will govern the existence and validity of IP rights, separately from the law that governs licensing or enforcement. A Singapore-based group that owns the AI model, licenses it to an Indian subsidiary. Additionally. Sells outputs globally may be able to claim that the model itself. as a software asset protected under Singaporean law. is the relevant IP, with the Indian subsidiary holding only a technology licence. That structure does not eliminate Indian law exposure, but it relocates the primary IP asset to a jurisdiction with a clearer protection regime.
The Arbitration and Conciliation Act (India's principal arbitration legislation) provides a well-developed mechanism for resolving cross-border IP and technology licensing disputes through arbitration. Parties with complex AI arrangements across multiple jurisdictions frequently choose to arbitrate disputes rather than litigate them, given the consistency and enforceability of arbitral awards under the New York Convention framework. Indian courts have been supportive of arbitration in commercial disputes, and institutional arbitration rules administered by bodies such as the Singapore International Arbitration Centre are regularly chosen for India-related technology contracts.
Regulatory dimensions compound the picture. The Securities and Exchange Board of India (SEBI) has indicated interest in AI governance for financial services applications, including algorithmic accountability requirements for AI systems used in trading and investment advice. The Reserve Bank of India (RBI) has issued guidance on the use of AI in regulated financial activities. Companies deploying AI in fintech, lending, or investment contexts in India must address both intellectual property ownership and regulatory compliance as linked questions. IP rights that cannot be enforced, or that are held in a structure that regulators consider non-compliant, provide limited commercial protection.
The National Company Law Tribunal (NCLT) and the broader corporate litigation system under the Companies Act 2013 are also relevant for AI businesses structured as Indian entities. Disputes over IP ownership between shareholders or between a company and its founders – common in AI startups – may be resolved through NCLT proceedings. The IP ownership question intersects with the corporate governance question of who controls the AI assets of the entity.
For a comprehensive view of how AI regulation and technology law apply across the full business lifecycle in India, our practice group provides integrated advice: AI and technology law services in India.
To discuss how cross-border AI intellectual property structures apply to your business across Asia-Pacific and Middle Eastern markets, contact us at info@ferrazwhitmore.com.
Strategic recommendations and the regulatory outlook
Given the current state of Indian law, a defensive strategy built on a single IP instrument is inadequate. Companies that rely exclusively on copyright registration for AI outputs. Alternatively, that assume patent protection will follow from AI-assisted innovation. Are likely to discover gaps in their protection at the worst possible moment. when enforcement is needed and a defendant raises a validity challenge.
A layered strategy is more resilient. Its components, applied in combination, address the doctrinal uncertainty without waiting for legislative or judicial resolution.
The first component is human authorship documentation. Companies should establish, at the point of creation, a contemporaneous record of the human decisions that shaped the AI output. This includes the prompt architecture, the selection criteria applied to candidate outputs, the editorial adjustments made by human staff, and the business judgment exercised in choosing one output over another. That record supports the argument that human authorship is present at a level sufficient to attract copyright protection. It also strengthens the inventive step narrative for patent purposes.
The second component is trade secret and confidentiality protection. Under Indian law, confidential information is protected through contract and through the general law of obligations. An AI system's model weights, training data sets, fine-tuning parameters, and prompt engineering are all capable of protection as trade secrets. Unlike copyright, trade secret protection does not depend on resolving the authorship question. It depends on the information being genuinely confidential and on reasonable steps having been taken to maintain that confidentiality. Digital services businesses should audit their technical and contractual measures and tighten them where gaps exist.
The third component is contract-based protection. Technology licensing agreements should assign all rights that may exist in AI outputs – however legally characterised – to the intended rights holder. They should contain representations about training data, indemnities against third-party IP claims, and clear provisions on what happens to AI-generated works if the agreement terminates. Employment contracts and contractor agreements for AI development staff should be reviewed against the same framework. The goal is to ensure that no AI output can vest in an individual employee or contractor rather than in the company.
The fourth component is monitoring the regulatory trajectory. India does not yet have AI-specific legislation comparable in scope to the EU's AI Act. However, the government has signalled an intent to develop an AI governance regime. Regulatory consultations have addressed algorithmic accountability, data localisation, and the responsibilities of AI platform operators. When India's AI regulatory regime takes shape, it is likely to address IP ownership, software liability, and training data use in ways that affect the strategies described above. Companies with significant AI exposure in India should engage with the regulatory process through industry bodies and maintain flexibility in their IP structuring.
The fifth component is jurisdictional arbitrage where lawful. A Singapore-incorporated holding company owning AI model assets, licensing them to an Indian operating entity, and enforcing IP rights from a jurisdiction with clearer protection offers a structure that many Asia-Pacific groups are already using. The structure must be commercially genuine and must comply with Indian transfer pricing rules, foreign exchange management legislation, and the tax provisions of the relevant bilateral treaty. The Indian tax authorities have increased scrutiny of IP holding structures. AI-specific structures require careful analysis before implementation.
Looking ahead, the Indian courts are likely to confront the authorship question within the next few years as AI-related commercial disputes increase in volume. The most likely judicial approach. drawing on purposive interpretation and the investment-incentive rationale – may extend protection to AI-assisted works where meaningful human direction can be demonstrated, while denying protection to fully autonomous AI outputs. That outcome would reward companies that have built contemporaneous authorship documentation and would penalise those that assumed protection without structuring their workflows accordingly.
The legislative trajectory is less predictable. Indian intellectual property legislation has historically evolved through targeted amendments rather than wholesale recodification. An amendment expressly addressing AI authorship and AI inventorship would resolve the current uncertainty but would also require political consensus on contested questions about the scope of protection. The identity of the rights holder. Additionally, the treatment of training data. The more likely near-term outcome is judicial guidance followed by targeted legislative adjustment, rather than a comprehensive AI intellectual property statute.
For businesses that cannot afford to wait for that clarity, the strategies outlined above provide the most durable available protection under current conditions. The opportunity cost of inaction – particularly for companies whose business model depends on exclusive exploitation of AI-generated content or AI-assisted inventions – is real and quantifiable.
Frequently asked questions
Q: Can an AI system be named as the author of a work under Indian intellectual property law?
A: No. Indian intellectual property legislation requires that an author be a human person. An AI system cannot hold authorship or copyright. The human who creates, directs, or substantially shapes the AI-generated output may assert authorship, but that claim requires demonstrating meaningful creative input beyond simply running a prompt.
Q: How long does it take to register a copyright or patent for an AI-assisted work in India?
A: Copyright registration in India typically takes several months from filing to certificate issuance, though registration itself is not mandatory for protection. Patent examination for AI-assisted inventions often extends beyond two years, partly because examiners frequently raise objections around inventive step and the human-inventor requirement. Applicants should factor in time for responding to office actions.
Q: Is it a misconception that training an AI on copyrighted data is automatically lawful in India?
A: Yes, this is a common misconception. Indian intellectual property legislation contains a fair-dealing provision, but it does not expressly authorise large-scale commercial AI training on copyrighted works. Courts have not yet settled whether such training constitutes infringement. Businesses relying solely on fair dealing for training data carry meaningful legal exposure until the position is clarified by legislation or judicial decision.
About Ferraz & Whitmore
Ferraz & Whitmore is an international law firm based in Lisbon, advising business clients across 46 jurisdictions. Our AI and technology law practice assists companies navigating intellectual property ownership, technology licensing, software liability, and algorithmic accountability across Asia-Pacific, the Middle East, and European markets. Engaging a lawyer in India with cross-border experience in AI regulation requires a team that understands both the local statutory environment and the international structuring options available to multinational groups. Our Senior Associate Anna Chen leads the firm's Asia-Pacific and Middle East practice, working with technology companies, investors, and institutional clients on AI-related IP strategy, cross-border regulatory compliance, and dispute resolution through arbitration and litigation. As an international law firm working across India and 45 other jurisdictions, Ferraz & Whitmore combines Portuguese civil law expertise with English common law tradition to deliver solutions that hold across multiple legal systems. The firm's technology practice includes practitioners with experience in proceedings before arbitral bodies including the Singapore International Arbitration Centre and in cross-border IP enforcement across Asia and the EU. To discuss your AI intellectual property position in India or across the region, contact us at info@ferrazwhitmore.com.
Disclaimer: This publication is provided for informational purposes only and does not constitute legal advice. The information herein should not be relied upon as a substitute for professional legal counsel tailored to your specific circumstances. Ferraz & Whitmore assumes no liability for actions taken or not taken based on the contents of this material. For advice regarding your particular situation, please contact info@ferrazwhitmore.com.