By Bhaskar Subramanian, Advocate, S&S Co. Advocates & Solicitors
Published 24 October 2026 · Law stated as on 19 October 2026 · Informational content, not legal advice: see our disclaimer
Generative AI has reached Indian courtrooms faster than Parliament has reached generative AI. Four strands of litigation, on authorship, training data, persona and branding, now give a reasonably clear picture of where Indian intellectual property law stands, even without a dedicated AI statute.
Key points
- Indian copyright law still insists on a human author; the Copyright Office has twice refused to let an AI system itself be named as author.
- The Delhi High Court's 2026 interim order in ANI v OpenAI treated the use of news content to train a model as arguably protected fair dealing, but the underlying suit on outputs continues.
- A string of personality rights orders, for Anil Kapoor, Jackie Shroff, Arijit Singh and Aishwarya Rai Bachchan, now protects name, voice, image and persona against AI cloning through interim injunctions.
- From 20 February 2026, intermediaries must label synthetically generated content and preserve identifying metadata under an amendment to the IT Rules, 2021.
- Ordinary trade mark and passing-off principles, including the Cadila test for deceptive similarity, apply unchanged to disputes over AI-chosen brand names.
- A business building on or deploying generative AI in India faces a set of overlapping rulings and rules rather than one statute, and should plan around all of them.
Why Generative AI Strains an Analogue Statute
The Copyright Act, 1957 was drafted for a world of human writers, composers and painters, and it shows. Trade mark law assumes a human mind choosing a name to build goodwill. Personality rights, to the extent India recognises them at all, grew out of defamation and privacy law aimed at human impersonators, not at a model that can generate a convincing voice clip in seconds from a few minutes of audio. None of these bodies of law anticipated a tool that can write a news report, paint an image, clone a voice or coin a brand name, at scale, on demand.
Parliament has not passed, and has not seriously proposed, a standalone AI statute. What exists instead is a set of court orders applying old statutes to new facts, together with one significant executive rule change aimed narrowly at synthetic audio-visual content. For a client asking what the law says about generative AI in India, the honest answer in October 2026 is that there are now enough decided points to draw a reasonably confident map, but the map has real gaps where no court has yet ruled.
That map divides into four questions that recur in practice: who, if anyone, owns what an AI tool produces; whether training a model on copyrighted material is itself an infringing act; what happens when a model's output overlaps with someone else's protected work, voice or face; and how trade mark and passing-off law responds when an algorithm, rather than a marketing team, throws up a brand name. Each has moved in the last two years, and the remainder of this piece works through them in that order before turning to the new labelling rules and their practical consequences.
Can a Machine Be an Author? The Section 2(d)(vi) Problem
Section 2(d)(vi) of the Copyright Act defines the author of a computer-generated literary, dramatic, musical or artistic work as the person who causes the work to be created. That clause was inserted by the 1994 amendment, long before anyone in Parliament was thinking about large language models, and it has become the fault line on which every Indian AI-authorship dispute has so far been decided.
The clearest illustration is the Suryast registration. In 2020, Ankit Sahni applied to register an image generated with the help of an AI painting tool called RAGHAV, first naming RAGHAV as sole author. The Copyright Office rejected that application on the footing that only a human can hold authorship. Sahni then refiled naming himself and RAGHAV as co-authors, and that version was registered in November 2020. Within roughly a year the Office issued a notice seeking to withdraw the registration, pointing to the unresolved question of whether an AI system can be a co-author at all, and commentators still disagree on whether the registration formally stands or has been set aside.
The Office took a firmer line in the DABUS matter. Dr Stephen Thaler sought registration for an artwork generated by his DABUS system, naming DABUS as author. Acting on a Delhi High Court direction in April 2026 to decide the long-pending application within eight weeks, the Registrar of Copyrights held, on 31 August 2026, that the image met the originality threshold under Section 13 but that DABUS could not be the statutory author under Section 2(d)(vi), since it is neither a natural nor a juristic person. The Registrar left Thaler free to refile naming himself, as the person who caused the work to be created, as author, treating DABUS as the tool rather than the author.
Read together, Suryast and DABUS point the same way. An AI tool can produce a copyrightable image or text, but Indian law will locate authorship in the human who directed, prompted or curated that output, not in the software. A business commissioning AI-generated content should document who gave the operative creative direction, since that person, not the tool and not necessarily the AI vendor, is where authorship will be anchored if the work is ever litigated.
Training on Copyrighted Material: The Fair Dealing Question
A separate and commercially larger question is whether feeding copyrighted material into a model during training infringes the rights in that material. India's leading answer so far comes from ANI Media Pvt Ltd v. OpenAI OpCo LLC, filed before the Delhi High Court in November 2024 as CS(COMM) 1028/2024. ANI alleged that OpenAI had used its news reports, without a licence, to train ChatGPT, and that the chatbot both reproduced its reporting and, on occasion, attributed fabricated statements to ANI.
OpenAI resisted on jurisdiction, arguing its servers sit abroad and that its terms point users to courts in San Francisco. In an interim order dated 24 July 2026, Justice Amit Bansal rejected that objection, holding that ANI's registered office is in India and that the allegedly infringing output reaches Indian users here, which is enough to found territorial jurisdiction regardless of where the servers are located.
On the merits, at the interim stage only, the court declined to grant ANI an injunction. It treated the storage and use of ANI's articles for training purposes as arguably falling within the fair dealing exception in Section 52(1)(a), which permits dealing with a work for private use and research. On ANI's separate complaint that ChatGPT's outputs reproduced its reporting, the court found overlapping facts and quotations but different sentence structure and framing, which it was not prepared to treat as substantial copying at the interim stage. It also declined to accept ANI's argument that the model had memorised its text verbatim, holding that what a model does or does not retain internally is a disputed factual question that needs evidence and trial, not something a court can resolve on an interim application.
Two things follow for now. First, this is an interim order on an injunction application, not a final judgment, and the suit itself continues; nothing here finally settles whether training on copyrighted news content is lawful in India. Second, the order does give the first judicial signal from an Indian court that fair dealing may extend to the mechanics of model training, which is a materially different reading from what some rights-holders had hoped for, and publishers who licence content to AI developers should treat this as the current baseline rather than as a settled defeat or victory for either side.
When AI Output Overlaps With Someone Else's Work
Separately from training, a model's output can infringe if it reproduces a substantial part of a specific protected work. Indian copyright law has always asked, for conventional works, whether the alleged copy reproduces the particular selection, arrangement and expression of the original, not merely facts or ideas drawn from it. Nothing in the ANI order displaces that test; the court simply found, on the material before it at the interim stage, that the overlaps it was shown did not cross that line.
That leaves open harder cases that Indian courts have not yet had to decide: a model reproducing a distinctive paragraph near-verbatim, regenerating a copyrighted illustration closely enough to count as a derivative work, or completing a copyrighted song in a way that reproduces its melody. The legal test for each is the ordinary one for substantial similarity, applied work by work, but no reported Indian decision has yet had to apply it to a generative output found to cross the line.
A related and distinct problem is attribution error, which ANI also raised: a model inventing a quotation or a claim and attributing it to a real publication or person. That sits closer to defamation and to the tort of passing off misrepresentation than to copyright, since the complaint is not that expression was copied but that a false statement was put in someone's mouth. Indian law has well-established defamation and passing-off remedies for exactly this kind of misattribution; what is new is only the mechanism, a model generating the false statement rather than a human writer inventing it.
For a business publishing AI-assisted content, the practical lesson from this strand of cases is to treat an AI draft the way an editor treats a trainee's draft: check quotations against the original source, verify that factual claims are not invented, and confirm that passages are not reproduced closely enough from an identifiable source to raise a substantial-similarity question, before publication rather than after a complaint arrives.
Faces, Voices and Catchphrases: Personality Rights Against AI Cloning
The most developed body of Indian case law on generative AI is not about copyright at all. It is a run of personality rights suits brought by well-known individuals against websites, merchandise sellers and AI platforms that used their name, face, voice or catchphrase without consent. India has no standalone personality rights statute; these cases are built from a mix of the tort of passing off, the right to privacy recognised under Article 21, elements of trade mark law, and in some suits the Copyright Act's performer's rights.
The sequence starts with Anil Kapoor v. Simply Life India & Ors, CS(COMM) 652/2023 before the Delhi High Court. In an order dated 20 September 2023, Justice Prathiba M. Singh restrained the defendants from using Kapoor's name, image, voice or his well-known catchphrase for merchandise, ringtones or AI-based alterations, and directed the takedown of infringing domain names. Jaikishan Kakubhai Saraf alias Jackie Shroff v. The Peppy Store & Ors followed in 2024, with Justice Sanjeev Narula granting an interim order on 15 May 2024 restraining a wide set of defendants, including an AI chatbot platform deploying Shroff's persona, from unauthorised use of his name, voice and image, and directing the Department of Telecommunications and the Ministry of Electronics and Information Technology to have the infringing links blocked.
The Bombay High Court extended the same reasoning to AI voice cloning in Arijit Singh v. Codible Ventures LLP & Ors, in an ex parte ad interim order in July 2024 restraining a large group of defendants, among them several AI platforms offering tools to convert text or audio into an AI-generated version of the singer's voice, from using his name, image or voice without consent. The Delhi High Court's order of 9 September 2025 in Aishwarya Rai Bachchan v. Aishwaryaworld.com & Ors, CS(COMM) 956/2025, before Justice Tejas Karia, went further still, restraining defendants from using her name, likeness or image through AI, deepfakes or face-morphing technology, and treating the misuse as harming both her commercial interests and her dignity.
Every one of these is an interim, ex parte order, not a final adjudication after trial, and each was fact-specific to the defendants named. But taken together they show Indian courts applying existing passing-off and privacy doctrine comfortably to AI cloning, without waiting for a dedicated statute, and they give any public figure, and by extension any business whose brand depends on a named spokesperson or founder, a workable, if not yet fully tested, template for approaching a court quickly when an AI clone surfaces.
Trade Marks and Passing Off in an Algorithmic Marketplace
Generative AI has also generated a wave of trade mark disputes, though these turn out to be ordinary trade mark problems wearing an AI-era coat. OpenAI's application to register ChatGPT has drawn a formal opposition in India from Flaxxi AI Private Limited, a Bengaluru entity asserting prior use of a similar mark. Google's application for Gemini has run into a challenge from Sun TV Network, which holds the long-standing Gemini TV mark, and X Corp's application for Grok has met a provisional refusal citing a conflict with an existing mark, Groke. None of these has yet been finally decided by the Trade Marks Registry or a court; they sit as live oppositions and refusals.
A more direct passing-off dispute is under way in Belagavi, where an Indian company, Anthropic Softwares Private Limited, which says it has traded under that name since 2017, has sued the American AI company Anthropic PBC in a commercial court, alleging that AI search tools and Google results now push the American company's name over its own and cause confusion among its customers. As of the most recent reporting, the American respondent had not appeared, and the court had issued fresh summons. No ruling on the merits exists yet.
What all of these disputes confirm is that an AI-era brand name is tested by exactly the same law as any other. Where two marks are confusingly similar, courts apply the established multi-factor test for deceptive similarity used in Cadila Health Care Ltd v. Cadila Pharmaceuticals Ltd, weighing the marks, the goods or services, the trade channels and the likely class of purchaser. A passing-off claim still needs goodwill, misrepresentation and resulting damage. That a name was suggested, chosen or popularised by an algorithm rather than a marketing department has no bearing on any of those elements; a company that lets a generative tool propose a brand name still carries the ordinary obligation to clear it against existing marks before adopting it.
For businesses, the lesson is unglamorous but important: AI-suggested names need the same trade mark clearance search as any other candidate name, and a large global AI company entering India under a name already used locally is not protected from an existing Indian trader's prior rights merely by its scale or by the fact that search engines now favour it.
The New Labelling Regime: Synthetically Generated Information
Where litigation has moved issue by issue, the executive has made one targeted rule change. On 10 February 2026, the Ministry of Electronics and Information Technology notified the Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026 as G.S.R. 120(E), amending the existing 2021 IT Rules, with effect from 20 February 2026, after a draft had been floated for comment in October and November 2025.
The amendment defines synthetically generated information, or SGI, as audio, visual or audio-visual content that has been artificially or algorithmically created or altered so that it appears authentic or indistinguishable from a real person or event, with carve-outs for routine editing such as colour correction and for clearly illustrative or educational material. Significant social media intermediaries must now obtain a declaration from a user uploading such content, verify that declaration using reasonable technical measures, and ensure the content carries a visible label and embedded, non-removable metadata identifying it as synthetic.
The amendment also tightens intermediary obligations generally, replacing the previous thirty-six-hour takedown window with a much shorter three-hour window under Rule 3(1)(d) for certain categories of flagged content, and making clear that acting to remove or disable access to non-compliant SGI, following the prescribed process, does not cost an intermediary its safe harbour under Section 79 of the Information Technology Act, 2000. Conversely, a platform that knowingly allows unlabelled or falsely declared synthetic content to circulate risks being treated as having failed its due diligence obligations, which would expose it to losing that safe harbour.
This is a narrow, platform-facing rule, aimed at labelling and metadata rather than at creating any new substantive right for the person depicted. It does not itself give Anil Kapoor or Aishwarya Rai Bachchan a cause of action; that still comes from the personality rights line of cases discussed above. What it does is give platforms a freestanding statutory reason to take synthetic content down quickly once notified, independent of whether the underlying use also happens to infringe a copyright, trade mark or personality right, and it is the first piece of Indian law that defines synthetic media as a distinct regulatory category in its own right.
What This Means for Businesses Building or Using AI Tools
A company adopting generative AI in its operations in India today is not working with a clean slate, but it is also not working with a single rulebook. Five separate bodies of law now bear on the same set of facts: copyright authorship and infringement, the fair dealing defence for anyone training or fine-tuning a model on third-party content, personality rights for any output touching a real person's name, voice or image, trade mark and passing-off law for any AI-suggested brand name, and the IT Rules labelling obligations for any platform hosting synthetic audio-visual content.
On authorship, the safest course remains to identify and document the human who exercised real creative judgment over an AI-assisted work, whether through detailed prompting, selection among outputs or substantial post-editing, since that is where Indian law will locate the copyright if the matter is ever contested, consistent with both the Suryast and DABUS outcomes. A business relying on an AI vendor's output should also check its licence terms for who is said to own that output, since contract can allocate rights between the business and the vendor even where the underlying copyright position is contested.
On training data, a business building or fine-tuning a model on scraped or licensed Indian content should treat the ANI interim order as encouraging rather than conclusive, and should still prefer licensed datasets where available, since the underlying suit on outputs, and the fair dealing question generally, remain open for final decision. On personality rights, any use of a real person's likeness, voice or persona, including for marketing, should be cleared by consent or licence; the cluster of 2023 to 2025 orders shows Indian courts moving quickly and broadly against unauthorised AI likeness use, usually within days of a suit being filed.
On platform obligations, any business operating a user-facing platform that could host AI-generated audio-visual content should already have built the labelling, metadata and verification workflow the 2026 amendment requires, given the short runway between notification and the 20 February 2026 effective date, and should treat a complaint about unlabelled synthetic content as carrying real urgency given the compressed takedown window.
None of this adds up to a single, coherent Indian law of generative AI, and none of the courts asked to rule on these questions has pretended otherwise. What exists is a set of interim orders, one Copyright Office decision under appeal pressure, and one targeted set of platform rules, each answering a narrow question while leaving the larger one, how far existing statutes stretch to cover machine-generated expression, for another day and very possibly for Parliament rather than the courts.
A client asking for certainty will not get it from this area of law in 2026. What a client can reasonably get is a working set of rules of thumb: keep a human author in the loop and on the record, prefer licensed training data where the budget allows it, clear any real person's likeness before using it, and build the labelling obligation into any platform before a regulator or a litigant has to point it out. That is not the same as a settled statute, but on the evidence of the cases decided so far, it is the position most likely to hold up as the remaining questions are litigated through to final judgment.
Frequently Asked Questions
If I generate an image or article using an AI tool, can I copyright it in India?
Indian law requires a human author, and the Copyright Office has refused to name an AI system itself as author in both the Suryast and DABUS matters. If you can show you directed the work through prompting, selection and editing, you can likely register it in your own name as the person who caused it to be created, but the AI tool cannot hold authorship.
Does training an AI model on my published articles or photographs infringe my copyright?
The Delhi High Court's 2026 interim order in ANI v OpenAI treated such training as arguably covered by the fair dealing exception for research and private use, but only at the interim stage, on an injunction application. The underlying suit continues, and the question is not finally settled under Indian law.
Someone has created an AI deepfake using my face or voice without my permission. What can I do?
Indian courts have granted swift interim injunctions in similar cases involving Anil Kapoor, Jackie Shroff, Arijit Singh and Aishwarya Rai Bachchan, usually within days of filing, based on passing off, privacy and, where relevant, performer's rights. A suit seeking an urgent ex parte injunction, naming the platforms and domains involved, is the established route.
Do the new IT Rules require every piece of AI-generated content to carry a label?
The amendment effective from 20 February 2026 requires labelling and metadata for synthetically generated information that is designed to appear authentic or indistinguishable from a real person or event, with exceptions for routine editing and clearly illustrative or educational material. It applies the strictest obligations to significant social media intermediaries, not to every individual creator.
Can I register a trade mark for a brand name that an AI tool suggested to me?
Yes, subject to the same clearance search and the same deceptive-similarity test, including the Cadila factors, that applies to any other proposed mark. The fact that an algorithm suggested the name does not relax the obligation to check it against existing registered and unregistered marks before adopting it.
Is there a dedicated Indian law for artificial intelligence that I should be checking?
Not yet. There is no standalone AI statute in India as of October 2026. The governing framework is a combination of the Copyright Act, 1957, ordinary trade mark and passing-off law, privacy jurisprudence under Article 21, and the 2026 amendment to the IT Rules on synthetically generated content, applied through a series of court orders rather than a single code.
References & Further Reading
The provisions, rulings and official documents below were relied on for this article, which states the law as on 19 October 2026. The position can change with amendments, rules and later decisions, so check the current text before acting on any of it.
- Copyright Act, 1957, Section 2(d)(vi) (author of a computer-generated work)
- Copyright Act, 1957, Section 13 (works in which copyright subsists) and Section 52(1)(a) (fair dealing)
- Information Technology Act, 2000, Section 79 (intermediary safe harbour)
- Information Technology (Intermediary Guidelines and Digital Media Ethics Code) Amendment Rules, 2026, G.S.R. 120(E), notified 10 February 2026, effective 20 February 2026
- ANI Media Pvt Ltd v. OpenAI OpCo LLC & Anr, CS(COMM) 1028/2024, Delhi High Court, interim order dated 24 July 2026 (Bansal, J.)
- Re: application of Ankit Sahni for registration of 'Suryast' naming RAGHAV as co-author, Copyright Office, registration granted November 2020, subsequent withdrawal notice
- Re: application of Dr Stephen Thaler for registration of a DABUS-generated work, Registrar of Copyrights order dated 31 August 2026, following Delhi High Court direction of April 2026
- Anil Kapoor v. Simply Life India & Ors, CS(COMM) 652/2023, Delhi High Court, order dated 20 September 2023 (Prathiba M. Singh, J.), 2023 SCC OnLine Del 6914
- Jaikishan Kakubhai Saraf alias Jackie Shroff v. The Peppy Store & Ors, Delhi High Court, interim order dated 15 May 2024 (Sanjeev Narula, J.)
- Arijit Singh v. Codible Ventures LLP & Ors, Bombay High Court, ex parte ad interim order, July 2024
- Aishwarya Rai Bachchan v. Aishwaryaworld.com & Ors, CS(COMM) 956/2025, Delhi High Court, ad interim order dated 9 September 2025 (Tejas Karia, J.)
- Cadila Health Care Ltd v. Cadila Pharmaceuticals Ltd, (2001) 5 SCC 73 (test for deceptive similarity)
- Anthropic Softwares Pvt Ltd v. Anthropic PBC, Commercial Court, Belagavi, passing-off suit, pending