AI-Generated Content and Copyright in India: Who Owns It, What Can Go Wrong, and a Practical Risk Map

An agency in Ahmedabad delivers a festive campaign for a skincare brand. The key visuals came out of Midjourney, the copy was drafted in ChatGPT and tightened by a writer, and the new logo was "started in an AI tool" before the designer cleaned it up. Six months later, in a funding round, the investor's lawyer asks one question: does the company actually own all of this?

Most questions about AI-generated content copyright in India end up there, and the answer is rarely a clean yes or no. It depends on how much a human contributed, what the tool's terms say, what the agency contract says, and whether the asset should have been a trademark in the first place. This guide covers the law as of October 2026, the Copyright Office's August decision, what is still pending, and a risk map for your team.

Quick answer

India has no AI-specific copyright rule yet. The Copyright Act, 1957 protects only original works (section 13), and for a computer-generated work it treats as author "the person who causes the work to be created" (section 2(d)(vi)). On 31 August 2026, the Registrar of Copyrights applied that clause to an application by Stephen Thaler: an AI system cannot be an author, but the human who caused the work can be, and an AI-generated image can still clear the originality bar. That is one administrative order on one set of facts, not a court ruling, and DPIIT's paper on AI outputs is still unpublished. In practice, documented human input strengthens your claim, and contracts and trademarks cover what copyright may not.

What the Copyright Act says about AI-assisted work

Four provisions do most of the work.

• Section 13 protects original literary, dramatic, musical and artistic works, films and sound recordings. Software is a literary work.
• Originality, per the Supreme Court in Eastern Book Company v. D.B. Modak (2008) 1 SCC 1, needs more than labour and money: some skill and judgement and a minimal degree of creativity. A low bar, but a bar.
• Section 2(d)(vi) makes the author of a computer-generated literary, dramatic, musical or artistic work the person who causes it to be created. The UK has a similar clause; the US does not.
• Section 57 gives the author moral rights: to claim authorship and to object to distortion that harms their honour or reputation. They assume a human author, as the Copyright Office pointed out when objecting to an AI author.

Ownership is a separate question. Under section 17, the author is the first owner, but an employer owns work made in the course of employment, and a client who pays for a photograph, painting, portrait, engraving or film owns it unless agreed otherwise. Other work commissioned from an agency or freelancer, including copy, code and most design work, stays with the creator until assigned in writing.

From SURYAST to DABUS: what the Copyright Office has done

In November 2020, the Copyright Office registered SURYAST, an artwork Ankit Sahni produced with an AI painting app called RAGHAV, naming both as co-authors. Within about a year, the Office issued a withdrawal notice asking about RAGHAV's legal status, pointing to the definition of author in section 2(d). Sahni replied that the Office could not review its own decision. No final decision has been published, and reports as recently as early 2026 said the entry was still on the register. The US Copyright Office refused the same work in December 2023.

The clearer signal came this year. Stephen Thaler, the American scientist behind the DABUS AI system, had applied in 2022 to register an image called A Recent Entrance to Paradise, naming DABUS as author. After the Delhi High Court directed the Office in April 2026 to decide the pending application, the Registrar of Copyrights passed a reasoned order on 31 August 2026. It held that:

• the image met the originality threshold, because generation by an algorithm does not by itself make a work unoriginal;
• DABUS is neither a natural nor a juristic person, so it cannot be an author;
• Thaler, who built and configured the system, supplied its inputs and started the process, caused the work to be created and could have been named as author;
• extending authorship to AI is a decision for Parliament.

The application failed only because Thaler declined the chance to amend it. Don't over-read the order, though. It is an administrative decision, open to appeal to the High Court, and the Registrar said expressly that it was not ruling on every form of AI-assisted creation. Thaler built his own system. A marketing executive typing a prompt into someone else's tool is in a weaker position, and no Indian authority has yet said whether that person "causes" the work. One point is settled enough to act on: never name an AI tool as author or co-author.

What DPIIT has said, and what it hasn't yet

DPIIT's committee on generative AI and copyright, set up in April 2025, split its work in two. Part 1 of the working paper, published in December 2025, deals with training and proposes a mandatory blanket licence with royalties paid through a central body. We cover it in our guide to AI training and the DPIIT licensing model.

According to Part 1 itself, Part 2 will decide whether AI outputs can be protected, who the author should be, whether moral rights apply and who is liable for infringing outputs. DPIIT said in December 2025 it expected Part 2 in about two months. As of 3 October 2026, it has not been published, and the Act has not been amended.

A February 2024 Rajya Sabha reply called existing law "well-equipped" to protect AI-generated works. The Registrar read that as covering works made with AI tools, not AI authorship.

Work going abroad faces a stricter rule. The US Copyright Office said in January 2025 that prompts alone do not make an author, and on 2 March 2026 the US Supreme Court declined to hear Thaler's challenge to the human-authorship requirement.

The other risk: output that looks like someone else's work

Ownership is one risk; infringement is the other. An image generator can produce something close to a known illustration or another brand's logo, and a chatbot can repeat passages it has seen. The Indian test comes from R.G. Anand v. Delux Films (Supreme Court, 1978): ideas are free, but if an ordinary viewer gets an unmistakable impression that one work is a copy of the other, it infringes. A machine making the copy changes nothing.

ANI Media v. OpenAI in the Delhi High Court is mostly about training, but its outputs finding matters. On 24 July 2026, Justice Amit Bansal refused ANI an interim injunction, taking the prima facie view that the ChatGPT responses ANI relied on were not a substantial reproduction of its articles and that memorisation or regurgitation had not been shown. ANI's complaint about invented stories attributed to it remains for trial. On ANI's appeal, a Division Bench issued notice to OpenAI on 15 September 2026 and listed the appeal for 5 December 2026. This is an interim view, not a final judgment.

Courts look at what the output reproduces, and if your team publishes an AI image that copies protected artwork, the claim lands on whoever published it. The terms of the major tools we checked make the user responsible for how outputs are used. Check for similarity before anything goes live.

A realistic AI face or voice resembling a real person raises personality rights, covered in our guide to deepfakes and personality rights. And under the IT Rules amended in February 2026, large platforms must ask users to declare realistic synthetic images, audio or video, and label them.

Tool terms and contracts decide more than the statute

What the tool terms give you

Most tools say you own what you create, but read the words. OpenAI's terms, effective 1 January 2026, assign you its rights in output "if any" and warn that other users may get similar output. Midjourney's terms, effective 27 May 2026, say you own your assets to the extent the law allows, but a company with more than US$1 million a year in gross revenue must be on a Pro or Mega plan to own them, and Midjourney keeps a broad, perpetual licence to your prompts and images.

A tool cannot hand you rights the law does not recognise, so "you own the output" is only as strong as the copyright behind it. Output anyone can regenerate almost identically is also hard to treat as exclusive.

What agency and freelancer contracts should say

This is where most problems actually get fixed. Under section 19, an assignment must be in writing and signed. If it states no period, it is deemed to last five years; if it states no territory, it is presumed to cover India only. An email saying "all IP belongs to the client" does not do the job. For AI-assisted deliverables, cover:

• Disclosure of which AI tools were used, and for which elements;
• Prior approval before AI is used on a logo, mascot or packaging artwork;
• Tool compliance: a warranty that each tool was used within its terms, including any paid plan needed for commercial ownership;
• Confidentiality: no unreleased client material in tools that train on inputs or publish outputs by default;
• Assignment of all rights in the deliverables, including AI-assisted elements to the extent rights exist, worldwide and for the full term;
• Records: working files, version history and prompt logs handed over with the finals;
• Fair warranties: a freelancer can promise their own contribution is original and not knowingly copied, but cannot promise copyright exists in raw AI output.

AI-made logos: copyright is uncertain, trademark is not

Logos cause a lot of avoidable trouble. A founder generates a mark, likes it and prints it on everything. Two things can go wrong.

First, copyright in a purely AI-generated logo is uncertain, and registering a logo as an artistic work has an extra step: under the proviso to section 45, the application must carry a certificate from the Registrar of Trade Marks that no one else has registered or applied for an identical or deceptively similar mark.

Second, and worse, the generator may have produced something close to an existing mark or artwork. Under section 11(3) of the Trade Marks Act, 1999, a mark cannot be registered if its use could be stopped under the law of copyright or passing off.

The useful point is that trademark protection does not depend on authorship. Under section 18, any person claiming to be the proprietor of a mark used or proposed to be used can apply. The questions are distinctiveness and conflict with earlier marks, not who or what drew it. So run a trademark search, have a designer rework the logo, take a written assignment from whoever made it, and file the trademark. Treat any copyright as a bonus.

A risk map for AI-assisted work

The "likely position" column is our reading of current law and the August 2026 order, not a settled rule.

Keep a record of the human contribution

If authorship is questioned, in a dispute, due diligence or a copyright application, the side with records usually wins.

1. Log prompts and settings: tool, version, date, prompts, seeds or parameters.
2. Keep the rejects and note why you chose one. Selection is evidence of judgement.
3. Keep working files: layered design files, tracked-change drafts, Git commits showing what a person changed.
4. Write a one-page contribution note per important deliverable: who did what, which tools, what was changed by hand.
5. Record your inputs, such as your own photographs or sketches, and confirm you hold rights in them.
6. Run a similarity check before publishing: reverse image search, plagiarism check, licence scan.
7. Save the tool's terms as they stood when you used it.

What to say, and not say, in a copyright application

• Do name the human author, and the employer or assignee as owner where relevant, with the no-objection certificate or assignment.
• Do describe the work accurately, and keep your contribution note ready in case the examiner asks how it was made.
• Don't name an AI tool as author or co-author. On the DABUS reasoning, expect a refusal.
• Don't describe AI-generated material as hand-drawn. A registration on wrong particulars can be challenged, and the High Court can rectify the register under section 50.
• Don't register a raw output you plan to change heavily. Register the final work.

The DABUS order left open whether AI use must be disclosed in the register; until there is a policy, be accurate. Our guide to copyright registration in India covers the process and documents.

How MYCrave can help

MYCrave Consultancy & Services works with founders, agencies and in-house teams on the ownership side of AI-assisted work:

• Copyright registration for final works with documented human authorship;
• Trademark search and filing for AI-assisted logos and brand names, so protection doesn't hinge on authorship;
• Contract review of agency, freelancer and client terms for AI disclosure, assignment and warranties;
• An IP audit of what your team has made with AI, what you are likely to own, and what to fix first.

Own the human part, and write it down

Nobody can yet say exactly where India will draw the line on AI authorship. Parliament may legislate, DPIIT's Part 2 may shift the debate, and a court may hear an appeal from the DABUS order. Wherever the line lands, your position will be judged on what you can show.

So keep the human part real and keep the record. Use AI for speed, put human judgement into the assets that carry your brand, register the trademark for anything customers know you by, and fix the contract before the invoice. Then you can answer the investor's question with documents instead of a shrug.

Frequently asked questions

Keep reading