India's creative industries were valued at about ₹2.5 trillion in 2024 and contributed about 0.73 per cent of GDP. Currently, they stand at a critical juncture as generative artificial intelligence (AI) threatens to redefine established intellectual property frameworks by existing outside it.

The emergence of AI systems capable of generating text, images, music and video by training on vast corpora of copyrighted works has exposed significant lacunae in India's statutory regime, particularly the Copyright Act 1957, which predates modern AI by decades.

This analysis examines the legal, economic and policy dimensions of AI's impact on India's creative economy. Three converging developments demand immediate attention: first, the December 2025 Department for Promotion of Industry and Internal Trade (DPIIT) Working Paper proposing a revolutionary mandatory blanket licensing framework; second, the landmark litigation in ANI Media Pvt Ltd v OpenAI Inc currently before the Delhi High Court; and third, the proliferation of AI-enabled deepfakes and voice cloning threatening the personality rights of Indian artists and performers.

Copyright Act 1957: The human authorship requirement

The primary problem arises out of the wording of Section 2(d) of the Copyright Act 1957. It defines 'author' exclusively in human terms. For literary, dramatic, musical and artistic works, the author is invariably a natural person - the composer, artist or photographer. Section 2(d)(vi) provides that for 'computer-generated' works, the author is 'the person who causes the work to be created'. This provision, inserted by the 1994 Amendment contemplating software-generated outputs, proves wholly inadequate for contemporary generative AI systems.

The statutory ambiguity creates three cascading failures. First, works generated autonomously by AI systems fall into a legal vacuum. When DALL-E generates an image responding to a minimal textual prompt, the question of who 'caused' the work to be created admits no clear answer. The Delhi High Court in Navigators Logistics Ltd v. Kashif Qureshi held that copyright cannot subsist in computer-generated compilations absent human creative involvement, a principle now tested by generative AI of far greater sophistication.

Second, the ownership provisions under Section 17 presuppose identifiable human authorship. When authorship becomes ambiguous, ownership becomes legally uncertain. Third, unlike the United Kingdom's Copyright, Designs and Patents Act 1988, which provides under Section 9(3) that the 'author' of a computer-generated work is 'the person by whom the arrangements necessary for the creation of the work are undertaken', Indian law contains no equivalent mechanism. The DPIIT expert committee constituted in April 2025 is actively examining whether legislative amendment is required to address this lacuna.

Section 52: The fair dealing inadequacy

Section 52 of the Copyright Act establishes a closed list of fair dealing exceptions permitting use without infringement for: private or personal use; research; criticism or review; reporting current events; and certain educational purposes. Unlike the United States' open-ended fair use doctrine under 17 USC § 107, which permits courts to consider any use potentially fair based on four statutory factors, India's approach is narrowly prescriptive.

The critical failure emerges when applied to AI training. Text and data mining (TDM) for machine learning involves systematic reproduction and storage of copyrighted works, activities squarely engaging the reproduction right under Section 14(a)(i). The DPIIT Working Paper explicitly acknowledges that Section 52 "cannot be utilised without demonstrating how an activity fits the prescribed categories" and that this lack of flexibility creates "significant legal uncertainty" for AI developers.

Parliamentary confirmation of this position came in December 2024, when the Minister of State for Commerce stated in response to a Lok Sabha question that "the Copyright Act, 1957 obligates the user of Generative AI to obtain permission to use their works for commercial purposes if such use is not covered under fair dealing exceptions provided under Section 52".

This effectively confirms that commercial AI training on copyrighted materials without authorisation constitutes prima facie infringement under Indian law.

Patents Act 1970 and IT Act 2000

The Patents Act 1970 creates parallel ambiguities. Section 6 requires patent applications to be made by "any person claiming to be the true and first inventor", language interpreted to require human inventorship. The US Federal Circuit in Thaler v. Vidal held that the Patents Act's use of pronouns 'himself' and 'herself' unambiguously requires natural persons as inventors. The UK Supreme Court reached the same conclusion in Thaler v. Comptroller-General, unanimously holding that 'DABUS is not a person, let alone a natural person'. The April 2025 Draft Computer-Related Inventions Guidelines now explicitly recognise AI as potentially patentable where 'technical effect' is demonstrated, following Ferid Allani v. Union of India (2020).

Section 79 of the Information Technology Act, 2000 provides 'safe harbour' immunity for intermediaries: online platforms that merely host or transmit third-party content without modification. Generative AI services fundamentally disrupt this framework. When an AI platform generates content in response to user prompts, it creates new outputs algorithmically rather than neutrally conveying content. The Ministry of Electronics and Information Technology has responded with draft amendments to the Intermediary Guidelines addressing 'synthetic media', mandating visible labelling and metadata preservation for traceability.

India's first generative AI copyright test

In November 2024, Asian News International (ANI) filed a suit before the Delhi High Court alleging that OpenAI used ANI's copyrighted news content, including subscriber-exclusive material, to train ChatGPT without authorisation. Justice Amit Bansal recognised the "landmark implications for copyright jurisprudence in the technological era" and framed four pivotal questions for determination: (i) whether storage of copyrighted data for AI training constitutes infringement; (ii) whether generating responses using copyrighted data constitutes infringement; (iii) whether such use qualifies as fair dealing under Section 52; and (iv) whether Indian courts have jurisdiction given servers are located in the United States.

The case has attracted extraordinary intervention. The Federation of Indian Publishers (representing Penguin Random House, Bloomsbury, Cambridge University Press and 80% of India's publishing sector), Digital News Publishers Association and major music labels including T-Series, Saregama and Sony Music have sought to join proceedings, arguing that the outcome will determine the viability of content licensing as a business model.

The voice cloning precedent

The Bombay High Court established India's first judicial precedent against AI voice cloning, granting an expansive injunction protecting singer Arijit Singh, whose voice had been synthesised without authorisation by AI platforms. The injunction extended across all media including physical, digital and metaverse platforms, covering voice models, voice conversion technologies, synthesised voices, digital avatars, deepfakes and GIFs. Critically, the Court established that protected attributes include voice, vocal style, vocal technique, vocal arrangements, mannerisms and signature.

The personality rights trajectory

This decision builds on a rapidly developing line of personality rights jurisprudence. In Anil Kapoor v. Simply Life India (September 2023), Justice Prathiba M Singh of the Delhi High Court delivered the first judgment explicitly addressing AI deepfakes. In January 2025, the Delhi High Court in Global Health Ltd v. John Doe extended protection to healthcare professionals. The celebrity deepfake epidemic has reached alarming proportions. Mukesh Ambani accounts for about 24.3% of analyzed deepfake financial scam cases, with projected losses of about ₹70,000 crore by 2025.

The United States relies on courts applying the four-factor fair use test under 17 USC § 107 to AI training disputes, with outcomes highly uncertain. In Andersen v. Stability AI, Judge William Orrick found that copyrighted works "may be contained in Stable Diffusion as algorithmic or mathematical representations", a significant holding that AI models can embody copyrighted works in transformed form. The US Copyright Office's March 2023 guidance established that human authorship remains constitutionally required, with AI-generated content excluded from registration unless a human exercised 'sufficient creative control' over expressive elements.

The EU Copyright Directive 2019/790 establishes a two-tier text and data mining framework. Article 3 provides a mandatory exception for research organisations with no opt-out mechanism. Article 4 permits commercial TDM subject to rights holders expressing reservation "in an appropriate manner, such as machine-readable means". The EU AI Act (Regulation 2024/1689) requires general-purpose AI providers to implement policies demonstrating copyright compliance.

The United Kingdom's Section 9(3) of the Copyright, Designs and Patents Act 1988 uniquely provides that for computer-generated works, "the author shall be taken to be the person by whom the arrangements necessary for the creation of the work are undertaken". Japan adopted the most innovation-friendly approach through Article 30-4 of its Copyright Act, permitting broad ingestion of copyrighted works for 'information analysis' including AI training, for both commercial and non-commercial purposes, with no opt-out mechanism.

On December 8, 2025, the DPIIT released Part I of its Working Paper on Generative Artificial Intelligence and Copyright, representing India's most significant policy intervention on AI and intellectual property. The eight-member expert committee examined five regulatory models: blanket exemptions; text and data mining exceptions; voluntary licensing; extended collective licensing; and statutory licensing. The committee rejected both pure exception approaches and pure licensing models, proposing instead a 'hybrid model' combining statutory licensing with collective management.

Key elements of this are as follows:-

1. Mandatory blanket licence: AI developers receive statutory authority to use any published copyrighted work for training, without requiring individual consent, provided they participate in the royalty system. Critically, rightsholders cannot opt out.

2. Copyright Royalties Collective for AI Training (CRCAT): A government-designated non-profit umbrella body comprising existing copyright societies (IPRS, PPL and others) would collect and distribute royalties.

3. Revenue-share royalty model: Flat percentage royalties on AI companies' global gross revenue from commercialised systems, determined by a Rate Setting Committee subject to triennial review and judicial oversight. The model applies retroactively.

This proposal is a distinctive 'third way' between the EU's opt-out mechanism and the US's case-by-case fair use determination. Creative industry stakeholders have broadly welcomed guaranteed compensation but express concern about inability to withhold works from AI training. Technology industry representatives have cautioned that blanket licensing could burden early-stage companies.

Asian Development Bank studies estimate creative employment encompasses approximately 40 million workers, roughly 8 per cent of total employment. Stanford research and NASSCOM found that 22-25-year-olds in AI-exposed jobs experienced 13% lower employment since 2022, coinciding precisely with ChatGPT's launch.

Sectoral vulnerabilities are significant. Voice cloning poses existential threats to the music industry, with AI tools replicating voices with up to 95% accuracy. Major music labels have joined the ANI proceedings, arguing that unregulated use of copyrighted works by AI adversely affects licensing agreements. AI image generators now openly market ability to create 'Madhubani, Warli, or Pattachitra' styles. While 605 Geographical Indications (GIs) are registered in India, GI protection covers geographic names but not underlying design patterns.

The policy debate has crystallised around 'Bollywood vs. Bangalore.' Creative industry stakeholders advocate licensing-based frameworks. MPA India MD Uday Singh warned that unrestricted AI use could undermine the incentive to create new works and erode copyright protection. Tech industry stakeholders push for innovation-enabling frameworks, with India filing 86,000+ AI patents between 2010-2025, ranking 5th globally in patent filings.

India stands at a critical juncture in intellectual property regulation. The convergence of generative AI with creative industries has exposed fundamental inadequacies in statutory frameworks designed for a pre-digital era. The choice is not between innovation and creator protection; a well-designed regulatory architecture can achieve both.

The DPIIT's hybrid licensing proposal represents a bold attempt to chart a distinctive path, rejecting both the EU's opt-out mechanism and the US's litigation-heavy uncertainty. The stakes for India's 40 million creative workers cannot be overstated. As the ANI v OpenAI litigation proceeds and the DPIIT consultation closes, decisions made in the coming months will shape whether AI becomes a tool for cultural flourishing or an instrument of creative displacement. The outcome will resonate not only within India but as a model, for better or worse, for the Global South as it confronts the AI revolution.

Sasmit Patra is a Rajya Sabha MP and Supreme Court advocate.

Arjit Bansal is a second-year B.A. LL.B. student at National Law University, Delhi.