IPR
Topic 89 AI Copyright
Topic 89 — Artificial Intelligence and Copyright
The intersection of artificial intelligence and copyright law represents the most contested IP issue of the 2020s. Two distinct questions dominate: (i) Can AI-generated works be copyrighted? (ii) Does AI training on copyrighted material constitute infringement or fair use? Indian copyright law, framed by the Copyright Act 1957, predates generative AI by seven decades. The pivotal definition is Section 2(d)(vi) — "in relation to any literary, dramatic, musical or artistic work which is computer-generated, the person who causes the work to be created" is the author. India is one of the few jurisdictions globally that even contemplates "computer-generated works" — alongside the UK CDPA 1988. The Delhi High Court is actively adjudicating ANI Media (P) Ltd. v. OpenAI Inc. & Anr. [CS(COMM) 1028/2024; 2024 SCC OnLine Del 8120] — judgment reserved 2025 — addressing whether AI training on copyrighted news content constitutes fair use under Section 52. Stephen Thaler v. Copyright Office (Del HC 2025) directed the Copyright Office to decide AI authorship claims within 8 weeks. The Federation of Indian Publishers, NDTV, Indian Express, and Hindustan Times have filed similar suits. The Ministry of Commerce constituted an expert panel in May 2025. This topic walks through every aspect of AI-copyright law — authorship, training data, fair use, ownership, infringement, and the evolving Indian framework.
1. The Two Core Questions
Q1 AUTHORSHIP Can AI works be copyrighted? | Q2 TRAINING Is training fair use? |
2. AI Authorship — The First Question
A. Section 2(d)(vi) — The Indian Framework
Section 2(d)(vi) Copyright Act "Definitions.— (d) "author" means,— ... (vi) in relation to any literary, dramatic, musical or artistic work which is computer-generated, the person who causes the work to be created." |
B. The Three Possible Authorship Approaches
Approach | Author | Indian Status |
|---|---|---|
AI as author | AI system itself. | NOT recognised — non-human cannot be author. |
Developer as author | Person/company who built AI system. | Possible interpretation. |
User/Prompter as author | Person who provides prompts and curates output. | Likely "person who causes work to be created" under 2(d)(vi). |
Joint authorship | Multiple parties. | Theoretical; not yet adjudicated. |
No copyright | Public domain. | US position; some Indian commentators advocate. |
C. The Indian Cases
📖 Stephen Thaler v. Copyright Office (Del HC 2025) Background — Stephen Thaler, US-based AI inventor, filed application before Indian Copyright Office for an artwork generated autonomously by his DABUS (Device for Autonomous Bootstrapping of Unified Sentience) AI system. The artwork: "A Recent Entrance to Paradise". Application pending since 2022. Indian Petitioner Position — Thaler argued: although artwork was generated autonomously by AI, he as developer and operator satisfies authorship requirement. Holding — Justice Tushar Rao Gedela (single bench) directed Copyright Office to: (i) Conduct hearing on the application. (ii) Conclude proceedings within 8 weeks from hearing date. (iii) Decide on AI-generated artwork copyright registration. Significance — First major Indian case directly addressing AI authorship. Significance: (i) Court did not decide on merits — directed Copyright Office to decide. (ii) Recognises issue raises "larger questions on authorship in the age of artificial intelligence". (iii) Mandates timely adjudication. (iv) International parallel — DABUS rejected by USPTO and UK IPO; UK Court of Appeal affirmed rejection. (v) Indian framework: Section 2(d)(vi) "computer-generated work" provides more open framework than US law. |
✅ The "Raghav" earlier case — 2021 In late 2021, an AI-generated artwork created by an AI named "Raghav" (developed by Indian artist Ankit Sahni) was initially approved by Indian Copyright Office for registration. However, this approval was subsequently CANCELLED on grounds: · No human author identified. · Section 2(d)(vi) requires "person who causes work to be created". · Pure AI generation without human intervention does not qualify. The Raghav case demonstrated: · Initial regulatory uncertainty. · Subsequent reversal toward human-authorship requirement. · Need for clearer statutory framework. · Foreshadowed the Thaler case proceedings. Lessons: Pure AI generation without substantial human creative contribution faces registration challenges in India. |
D. The "Person Who Causes" Question
Section 2(d)(vi)'s "person who causes the work to be created" is the central interpretive question:
- User who provides creative prompt — likely sufficient for authorship.
- Curator who selects and refines output — likely sufficient.
- Developer of generic AI tool — uncertain.
- Pure passive prompter ("create something") — likely insufficient.
- Multiple contributors — joint authorship possible.
3. AI Training Data — The Second Question
A. The Section 52 Fair Dealing Question
Section 52 Copyright Act — Acts not constituting infringement "(1) The following acts shall not constitute an infringement of copyright, namely:— (a) a fair dealing with any work, not being a computer programme, for the purposes of— (i) private or personal use, including research; (ii) criticism or review, whether of that work or of any other work; (iii) the reporting of current events and current affairs, including the reporting of a lecture delivered in public. ..." |
B. ANI Media v. OpenAI — The Watershed Case
📖 Ani Media (P) Ltd. v. Open AI Inc. & Anr., CS(COMM) 1028/2024; 2024 SCC OnLine Del 8120 (Del HC) Facts — In November 2024, Asian News International (ANI) sued OpenAI before Delhi HC alleging: (i) OpenAI used ANI's copyrighted news content to train ChatGPT (LLM) without authorization. (ii) ANI material accessible only to subscribers; no authorization given. (iii) OpenAI republished material via ChatGPT. (iv) OpenAI also falsely attributed news to ANI causing reputational harm. OpenAI Defense — OpenAI submitted: (i) ANI website (www.aninews.in) blocklisted in October 2024. (ii) Domain excluded from future training. (iii) No interim injunctions granted in similar US/Canada/Germany cases. (iv) Use is transformative. Court Order (19 November 2024) — Delhi HC framed four key questions for adjudication: 1. Whether storage by OpenAI of ANI's data (news content protected under Copyright Act 1957) for training ChatGPT amounts to copyright infringement? 2. Whether OpenAI's use of ANI's copyrighted data to generate user responses amounts to copyright infringement? 3. Whether OpenAI's use qualifies as "fair use" under Section 52? 4. Whether Indian courts have jurisdiction (servers in USA)? Amici Curiae Appointment — Court appointed amici curiae to assist due to lack of precedents and complexity. Status — Judgment reserved 2025; pending decision. Significance — Watershed case for Indian AI-copyright law: (i) First major judicial scrutiny of generative AI in India. (ii) Will set precedent for AI training data rights. (iii) Will define jurisdiction for cross-border AI services. (iv) Implications: Federation of Indian Publishers, NDTV, Indian Express, Hindustan Times have filed similar actions. |
C. The Section 52 Fair Dealing Test
✅ Whether AI training qualifies as fair dealing — the key arguments AGAINST fair dealing (plaintiffs): · Section 52 list is EXHAUSTIVE per Super Cassettes Industries v. Chintamani Rao (Del HC). · AI training is commercial — fails private/personal use test. · Not for criticism/review. · Not for reporting events. · Use enables competition with original (LLM outputs replicate journalism). · Lacks transformative purpose. FOR fair dealing (defendants): · Use is transformative — building AI is different purpose. · Cambridge Press Syndicate v. B.D. Bhandari (Del HC) recognised transformative use. · Storage temporary; not displayed verbatim. · No market harm — different purpose. · International alignment — US permits. · Educational/research benefit. The DPIIT 2024 position rejects blanket fair use defence for commercial-scale AI training. EU AI Act 2024 recognises Text and Data Mining (TDM) exception with opt-out. India has no equivalent provision. |
4. International Comparisons
Jurisdiction | Authorship | Training Data |
|---|---|---|
USA | Human authorship required (Thaler rejected by US Copyright Office; affirmed by federal courts). | Andersen v. Stability AI; Authors Guild v. OpenAI; Bartz v. Anthropic — pending; Alter v. OpenAI 2024. |
UK | Computer-generated works (CGW) recognised under CDPA 1988 Section 9(3) — author is person who undertook arrangements. | No specific TDM exception for commercial use. |
EU | Human authorship required. | EU AI Act 2024 + DSM Directive 2019 — TDM exceptions with opt-out. |
China | AI-generated works increasingly recognised in Chinese courts (Filmwise case). | Permissive approach. |
Australia | Following UK approach. | Copyright Act review pending. |
Japan | AI-generated works generally not copyrightable (no human creator). | Article 30-4 — TDM permitted. |
India | Section 2(d)(vi) "computer-generated work"; Thaler Del HC pending decision. | ANI v. OpenAI pending; Section 52 fair dealing question. |
5. AI-Generated Output Ownership
A. Categories of AI-Generated Works
Category | Example | Likely Indian Treatment |
|---|---|---|
Pure AI generation | AI generates without human prompt. | NOT copyrightable (no human author). |
Minimal prompt | User: "Create art" | Likely NOT copyrightable. |
Detailed prompt + curation | User provides specific prompt + selects from outputs. | Likely copyrightable — user is "person who causes". |
AI-assisted (human led) | Human creates; AI helps. | Copyrightable to human creator. |
AI-modified existing work | AI modifies copyrighted work. | Derivative work issues; original copyright + AI contribution. |
B. Practical Ownership Strategies
✅ Strategies for AI-generated work ownership 1. Document creative input — preserve prompts, drafts, iterations, selections. 2. Demonstrate human creative contribution — even if AI generates, human selection/curation matters. 3. License terms with AI tools — many AI tool ToS specify ownership. 4. ChatGPT, DALL-E commercial use — review specific licensing terms. 5. For critical works, prefer AI-assisted (clearly human-led) over AI-generated. 6. For artistic works, document creative process beyond pure AI generation. 7. For literary works, ensure substantial human authorial contribution. 8. For software, AI-generated code raises distinct issues — Section 17 first ownership. |
6. Liability for AI Infringement
A. Categories of AI-Related Infringement
- Training infringement — AI trained on copyrighted material.
- Output infringement — AI generates substantially similar to copyrighted work.
- Attribution infringement — AI falsely attributing to copyright holder.
- Voice/likeness infringement — AI replicating performers/personalities.
- Derivative work infringement — AI deriving from copyrighted work.
B. Potential Defendants
Party | Liability Theory |
|---|---|
AI developer/owner | Direct infringement; secondary liability for outputs. |
User who prompts | Direct infringement if prompts request infringing content. |
Platform deploying AI | Intermediary liability under IT Act Section 79; potentially limited. |
Cloud service provider | Hosting AI training; secondary liability theory. |
Consumer of AI output | If subsequently uses output infringing way. |
C. The Section 79 Safe Harbour
- IT Act Section 79 — intermediary safe harbour.
- Conditions: due diligence; takedown on actual knowledge.
- AI platforms argue they are intermediaries.
- Plaintiffs argue AI platforms are publishers, not intermediaries.
- Question yet to be definitively settled in India.
7. The 2025 Expert Panel
A. Constitution and Mandate
In May 2025, Ministry of Commerce & Industry constituted 8-member expert panel:
- Mandate: evaluate Copyright Act 1957 adequacy for GenAI.
- Address authorship for AI-generated works.
- Define AI-generated content.
- Clarify ownership and accountability.
- Section 52 fair dealing reform.
- Coordination with international developments.
B. Expected Recommendations
Based on leaked deliberations (June 2025):
- New Chapter XII-A specifically for AI-generated content (possible).
- Defining "AI-generated work" statutorily.
- Training data licensing framework.
- Coordination with EU AI Act and UK approaches.
- Expert panel report expected late 2025/early 2026.
C. Implementation Timeline
✅ Expected legislative timeline Stage 1 — Expert panel report (late 2025). Stage 2 — Government consideration (2026). Stage 3 — Draft amendments (2026). Stage 4 — Public consultation. Stage 5 — Parliamentary consideration. Stage 6 — Enactment. Coordination with: · ANI v. OpenAI judgment. · Stephen Thaler decision. · International developments. · Industry consultations. |
8. Strategic Considerations
✅ For content creators — eight points Document creative process for AI-assisted works thoroughly. Watermark and metadata for AI training opt-outs. Monitor AI-generated content for unauthorized derivative use. For substantial creative input through prompts and curation, claim authorship. For pure AI generation, recognize copyright uncertainty. For training data licensing, coordinate with publisher associations. For ANI v. OpenAI outcome, monitor implications for sector. For commercial AI tools, review license terms carefully. |
✅ For AI developers — eight points For training data, obtain authorization or use clearly licensed sources. For commercial AI, document training data sources comprehensively. For DPIIT compliance, align with 2024 advisory positions. For Indian deployment, comply with MeitY 2024 advisory. For label requirements, label AI-generated content. For IT Act safe harbour, implement due diligence and takedown procedures. For litigation risk, monitor pending cases and prepare defenses. For India-specific, prepare for Section 52 fair dealing arguments. |
🎯 EXAM POINTERS — TOPIC 89
|