Author: Divyanshi Hazari Yadav, Indore Institute of Law
To the Point
Indian law protects three different interests that generative AI collides with: the copyright in works used to train a model, the copyright (if any) in what the model produces, and the personal identity of the people whose faces and voices are cloned. Courts have moved fastest on the third, granting sweeping injunctions to celebrities against deepfakes. On the first, the Delhi High Court’s interim order of 24 July 2026 in ANI v. OpenAI leans towards AI developers, though the appeal is pending. On the second, the statute still has no clear answer. This article argues that judge-made protection has outrun Parliament and that a statutory framework is now necessary.
Use of Legal Jargon
The Proof
1. The statutory and constitutional foundation
The law that applies to AI is a patchwork of older statutes. Nothing in it was drafted with generative models in mind.
• Copyright Act, 1957. Section 13 protects original literary, dramatic, musical and artistic works. Section 17 makes the author the first owner. Section 2(d)(vi) names, for computer-generated works, “the person who causes the work to be created” as the author. Section 51 defines infringement and Section 52 lists exceptions. Sections 38 to 38B protect performers’ rights, including moral rights, which matter for voice cloning.
• Constitution of India. Article 21 (life, dignity and privacy) is the root of personality rights, while Articles 19(1)(a) and 19(2) frame free speech limits such as parody and satire.
• Trade Marks Act, 1999. Celebrity names registered as marks gain statutory protection, and passing off covers false endorsement.
• Information Technology Act, 2000. Sections 66C (identity theft), 66D (cheating by personation), 66E (privacy violation), 67 and 67A (obscene content), together with Section 79 and the IT Rules, 2021, regulate platforms. The Rules require expeditious removal of impersonating or morphed intimate content.
• Bharatiya Nyaya Sanhita, 2023. Provisions on defamation (Section 356) and cheating by personation (Section 319) are used against deepfake misuse.
• Digital Personal Data Protection Act, 2023. A face or voice that identifies a person is personal data, so processing it without a lawful basis can attract obligations on the data fiduciary.
2. Issue One: Is training an AI model on copyrighted content infringement?
Training requires copying works into datasets, so the question is whether that copying is excused. India has no dedicated TDM exception. Section 52 is a closed list of fair dealing categories, unlike the open-ended U.S. fair use doctrine.
The Delhi High Court addressed this on 24 July 2026 in ANI Media Pvt. Ltd. v. OpenAI OpCo LLC (CS(COMM) 1028/2024). The suit was filed in November 2024. ANI alleged that ChatGPT was trained on its news reports without a licence and reproduced them in outputs. As reported, the Court made four key findings at the interim stage:
1. Jurisdiction. OpenAI argued its servers and training were outside India. The Court rejected this, relying on Neetu Singh v. Telegram FZ (2022), which held that territoriality cannot be applied rigidly in the age of cloud computing.
2. No memorisation. The Court was not persuaded that OpenAI permanently stores training data in order to regurgitate it. On the samples examined, ChatGPT’s responses were not a substantial reproduction of ANI’s articles.
3. Fair dealing. At the prima facie stage, training use was held to fall within the fair dealing exception in Section 52(1)(a)(i).
4. Interim relief refused. ANI failed to show a prima facie case, and an injunction would cause irreparable injury to OpenAI and the public at large. The Court also noted that OpenAI had stopped using ANI content for training after blocking its crawlers.
This is an interim order, and a finding of prima facie fair dealing is not a final declaration that AI training is lawful. On 15 September 2026, a Division Bench issued notice on ANI’s appeal, and the main suit continues on merits. Two amici curiae and six intervenors, including publishers’ and music industry bodies on one side and technology bodies on the other, assisted the Court. That shows how widely the outcome will be felt.
Parallel to the courts, the government’s own committee takes a different view. The DPIIT Working Paper on Generative AI and Copyright (Part 1, December 2025) noted that unauthorised training may amount to infringement under Section 51. It proposed a hybrid model: a mandatory blanket licence for all lawfully accessed works, with statutory royalties collected centrally and shared with rights holders (the “one nation, one licence, one payment” idea). Nasscom dissented in favour of a TDM exception with opt-out. The courts and the policy paper therefore point in different directions.
3. Issue Two: Who owns AI-generated output?
Three positions are possible: the user, the developer, or no one.
• Originality. In Eastern Book Company v. D.B. Modak (2008) the Supreme Court required a “modicum of creativity” and rejected both the labour-only and the pure-creativity extremes. A work produced with minimal human input struggles to meet this test.
• Section 2(d)(vi). The phrase “person who causes the work to be created” could favour the user who prompts the system, or the developer who built it. It was written for earlier computer-generated works, and a prompt may be too thin an input to count.
• Comparative position. U.S. courts insist on human authorship. In Thaler v. Perlmutter (D.C. Cir., 2025) a work created autonomously by an AI could not be registered. India has no equivalent ruling, and the Copyright Office’s reported registration of an AI co-authored work, followed by a move to withdraw it, shows the uncertainty.
A sound working rule is that where a human exercises real creative control (selecting, arranging, iteratively editing), the human is the author. Where output is purely autonomous, protection is doubtful. The DPIIT committee has signalled further guidance on AI outputs, but until then ownership is best settled by contract.
4. Issue Three: Deepfakes, voice cloning and personality rights
India has no statute on personality rights. Courts have built the doctrine from Article 21, passing off, trademark law, performers’ rights and the right of publicity. Since 2022 the Delhi High Court has become a preferred forum for public figures, and its orders have grown steadily broader. Common features are John Doe relief against unknown infringers, takedown directions to platforms within fixed windows (for example 72 hours in the Ramdev order of February 2026), and requirements to disclose subscriber details.
Two features matter for AI. First, the orders now name AI chatbot platforms and voice-cloning services as defendants. Second, they are increasingly technology-neutral, covering generative AI, face morphing and any future medium, so they do not need updating with each new tool.
The limits are real. Parody, satire, news and biography are protected by Article 19(1)(a), so courts must distinguish exploitation from expression. Most orders are interim and ex parte, and their reach over foreign platforms depends on enforcement. Ordinary people who are not celebrities rarely have the means to seek such relief, which creates an inequality of access that the law has not yet addressed.
Abstract
Generative AI raises three distinct legal questions in India: whether training on copyrighted works is infringement, who owns machine-generated output, and how individuals can stop deepfakes of their likeness and voice. This article examines each under the Copyright Act, 1957, the Constitution, the Trade Marks Act, 1999, the Information Technology Act, 2000 and recent judicial decisions. It reviews the Delhi High Court’s interim order of 24 July 2026 in ANI Media v. OpenAI, which found prima facie fair dealing in training and no proven memorisation, and notes the pending appeal. It sets this against the DPIIT’s proposed mandatory licensing model. It then traces the personality rights line of cases from Titan Industries to the 2026 orders of the Delhi and Bombay High Courts. The article concludes that the courts have protected identity effectively but that training and authorship remain unsettled, and recommends legislative clarity on three points: a licensing or TDM framework, a rule on authorship of AI output, and a statutory personality right with free speech safeguards.
Case Laws
A. Copyright and AI
1. ANI Media Pvt. Ltd. v. OpenAI OpCo LLC, CS(COMM) 1028/2024, Delhi High Court (24 July 2026). Interim injunction refused. Jurisdiction upheld. No prima facie memorisation shown. Training use treated as fair dealing under Section 52(1)(a)(i) at the interim stage. Appeal notice issued on 15 September 2026.
2. Neetu Singh v. Telegram FZ (Delhi High Court, 2022). Held that conventional territoriality cannot be applied strictly to cloud-hosted data. Relied on in ANI to take jurisdiction over a foreign AI company.
3. Eastern Book Company v. D.B. Modak, (2008) 1 SCC 1. Established the “modicum of creativity” test for originality, central to deciding whether AI-assisted works are protectable.
4. Thaler v. Perlmutter (U.S. Court of Appeals, D.C. Circuit, 2025). Persuasive authority that copyright requires human authorship.
B. Privacy and personality rights
5. R. Rajagopal v. State of Tamil Nadu, (1994) 6 SCC 632. Recognised the right to privacy under Article 21 and its limits in respect of public records and public figures.
6. K.S. Puttaswamy v. Union of India, (2017) 10 SCC 1. Privacy is a fundamental right. It is the constitutional anchor for control over one’s image and data.
7. Titan Industries Ltd. v. Ramkumar Jewellers (Delhi High Court, 2012). Defined the right of publicity as the right to control commercial use of one’s identity.
8. Amitabh Bachchan v. Rajat Nagi (Delhi High Court, 2022). An early blanket John Doe order protecting name, image, voice and persona.
9. Anil Kapoor v. Simply Life India (Delhi High Court, 2023). Restrained AI morphing, deepfakes and misuse of voice and catchphrases.
10. Jackie Shroff v. The Peppy Store (Delhi High Court, 2024). Addressed AI chatbots and misuse of name and speech style.
11. Arijit Singh v. Codible Ventures (Bombay High Court, 2024). Reported as the first AI voice-cloning injunction, invoking personality and performers’ rights.
12. Aishwarya Rai Bachchan v. Aishwaryaworld.com (Delhi High Court, 2025). Ex parteinjunction against generative AI deepfakes and unauthorised merchandise.
13. Sonakshi Sinha v. Character Technologies Inc. (Delhi High Court, March 2026).Technology-neutral injunction that covers AI chatbots, generative AI and face morphing, with chatbot operators as defendants.
14. Kajol v. Amazon and others (Delhi High Court, February 2026). Interim protection based on trademark, personality and passing off, with takedown of deepfakes against platforms including AI companion services.
15. Baba Ramdev (Delhi High Court, February 2026). Takedown of AI-generated and false endorsement content within 72 hours; misuse of persona treated as misappropriation and passing off.
16. Preity Zinta (Bombay High Court, June 2026). Leave granted to sue Google, Meta and others over AI deepfakes and chatbot personas, under Clause XII of the Letters Patent on the basis of reputational harm in Mumbai. Interim stage only.
Conclusion
The Indian response to generative AI is being written by judges, case by case. In personality rights, the result has been protective and fast: the Delhi and Bombay High Courts now treat deepfakes and voice clones as actionable misappropriation, and their technology-neutral orders are likely to age well. In copyright, the ANI v. OpenAI order gives developers some comfort, since it accepts, at least prima facie, that training is not infringement where no regurgitation is shown. It is not the last word. The appeal is pending, the evidence in the case was limited to samples, and the DPIIT’s licensing proposal reflects a very different policy view.
Three reforms would reduce this uncertainty. First, Parliament should choose between a statutory licence and a TDM exception with opt-out, and not leave the question to interim orders. Second, Section 2(d)(vi) should be clarified to state when AI output is protected and who owns it. Third, personality rights should be codified, with express carve-outs for parody, news and criticism, and a simple remedy for ordinary victims. Until then, creators should protect themselves by contract, platforms by prompt takedown, and developers by licensing and by avoiding verbatim output.
FAQs
1. Did the Delhi High Court declare AI training legal in India? No. The order of 24 July 2026 is an interim order that refused an injunction, finding prima facie fair dealing. The main suit is still pending and ANI’s appeal has been admitted for hearing.
2. Who owns content created by ChatGPT or similar tools in India? The law is unsettled. Section 2(d)(vi) refers to the “person who causes the work to be created”, but it has not been tested for generative AI. A human who makes real creative choices has the strongest claim. Use contracts and platform terms to settle ownership.
3. Can I be sued for training an AI model on public web content? Possibly. Public availability is not a licence. The ANI ruling helps developers, but it was decided on its own facts at the interim stage, and the DPIIT proposes mandatory licensing with royalties.
4. Is there a law that specifically protects against deepfakes in India? There is no single deepfake statute. Victims rely on personality rights, passing off, the IT Act (Sections 66C, 66D, 66E, 67), the IT Rules, 2021 and provisions of the BNS, 2023.
5. Do personality rights protect only celebrities? In principle, any person can claim them under Article 21. In practice, most reported orders favour public figures with established goodwill, and ordinary victims rely mostly on the IT Act and criminal law.
6. Is every AI parody or fan edit illegal? No. Satire, parody, commentary and news enjoy protection under Article 19(1)(a), and courts distinguish these from commercial exploitation, false endorsement or obscene content.
