Charting India’s AI–IP Playbook: Innovation, Rights and National Advantage

Detailed Summary

Theme	Key Points
Background & Credibility	40 years in civil service, PhD (IIT Delhi), law degree (DU), certifications (Manchester, Harvard), author of IPR Bride & Competition Groom, Privacy Right & Data Protection, and a forthcoming book on AI.
AI & IPR – Why It Matters	• AI‑driven tools can accelerate IPR‑related processes (e.g., patent prior‑art searches). • Big‑data’s 5 Vs – Velocity, Volume, Variety, Value, Trustworthiness – frame the scale of IPR data.
Design Principles for AI in IPR	• AI should augment—not replace—human expertise. • Framework must protect innovation, avoid bias, prevent discrimination, and respect privacy.
Upsides of AI	• Multiplier‑effect technology, democratising innovation. • Applications in healthcare (diagnostics, trial design). • Alignment with India’s “fifth industrial revolution” vision and PM Modi’s call for technology democratization.
Downsides / Risks	• Surveillance amplification, cyber‑crime (AI‑enabled attacks, IP‑injection). • Copyright tensions: generative AI “vampirical” use of existing works.
Regulatory Landscape	• Existing statutes (IT Act, IT Rules, Data Protection Bill) address parts of the problem but are insufficient for AI. • Need for a techno‑legal approach rather than a purely legal one (citing Minister Ashwini Vishnaw).
Safety Pillar of India AI Mission	• Development of AI tools for deep‑fake detection, bias mitigation, model “unlearning”.
Five Broad Principles for an AI‑IP Regime	1. TRIPS‑public‑interest override – public interest trumps WTO‑mandated TRIPS provisions when necessary. 2. Patient‑first – health‑related patents must yield to patient needs. 3. Public‑good supremacy – private IP rights bow to societal benefit. 4. Balanced competition – IPR protection co‑exists with antitrust enforcement. 5. “Festina lente” – legislate cautiously and incrementally.
Closing	Thanked the audience and handed over to the panel.

---

2. Panel Introduction & First Question

Moderator (Prof Shukla) set the agenda: 34 minutes total, rapid round‑robin to capture each panelist’s view on “major IP‑related challenges for AI training and AI‑generated content.”

2.1. Anne E. Robinson – IBM

• Open‑source as foundational – IBM’s “Granite” family of foundation models are positioned as the most transparent globally, emphasizing trust, auditability and collaboration.
• Balancing openness & downstream protection – Open‑source accelerates innovation (start‑ups can build without reinventing wheels) but must be coupled with IP safeguards to prevent a “free‑for‑all.”
• Policy recommendation – India’s IP framework should enable open‑source use while embedding downstream IP rights, turning openness into a competitive advantage.

2.2. Rai Rajani Vindokumar – Chubb

• “Ownership‑value‑link problem” – Without clear ownership, ROI calculations for AI projects become impossible.
• Quality trap – Synthetic data may produce noisy or stale models; rapid shifts in consumer behavior (pre‑/post‑COVID, cyber‑attacks) erode model relevance.
• Cost & risk – High model‑training and compute expenses, plus IP‑injection attacks (reverse‑engineering prompts) create legal‑risk overheads.
• Bottom line – Quantifying AI’s economic value hinges on explicit IP ownership and robust data quality controls.

2.3. Sabeen Malik – Rapid7

• Enforcement gap – Legal recognition alone insufficient; technological enforcement (watermarking, provenance tracking, automated infringement detection) is critical.
• Co‑branding of law & cybersecurity – Calls for a joint effort between legal teams and security researchers to develop tools that can attribute ownership of AI‑generated artefacts.
• Future frontier – Developing practical watermarking & provenance standards in collaboration with the U.S. and India.

2.4. Nitin Seth – Incedo

• Data abundance paradox – AI moves the world from scarce to abundant data; traditional notions of ownership become fuzzy.
• Three‑layer model:
  1. Macro (Gen‑AI) layer – Public‑domain, hard to restrict.
  2. Enterprise layer – Proprietary data/models; a competitive differentiator (the “pinch of salt”).
  3. Individual layer – Personal data privacy; needs global standards and clear consent mechanisms.
• Policy implication – Separate regulatory treatment for each layer; focus on protecting enterprise‑level IP while establishing individual‑data privacy norms.

2.5. Dr Shardul S. Shroff – Shardul Amarchand Mangaldas & Co

• Current legislative debate – Discussed the DPIT‑proposed mandatory blanket licence for AI commercialisation vs. NASSCOM’s call for a text‑and‑data‑mining (TDM) exemption.
• Key legal uncertainties:
  • Fair dealing vs. fair use (pending Delhi High Court decision in ANI v OpenAI).
  • Ownership & liability of AI‑created works.
  • Personality‑rights extensions for AI‑generated likenesses.
• Recommendation – Enact a statutory TDM exception (opt‑out right) to avoid stifling AI innovation, while preserving creators’ future‑income expectations.

2.6. Jennifer Mulvenny – Adobe

• Creative‑sector focus – Artists rely on Adobe tools; they need control over training data and rights over AI‑generated outputs.
• Opt‑out mechanisms – Adobe already offers content‑credential provenance that lets users declare “do‑not‑train” preferences; calls for legal enforcement of such preferences.
• Missing “impersonation right” – Proposes a new statutory right to address unauthorised AI‑driven likeness replication.

---

3. Follow‑up Discussion – Practical Concerns & Institutional Support

3.1. Startup & Small‑Business Uncertainty (Jennifer & Rajani)

• Legal clarity gap – Start‑ups lack certainty on IP exposure when building AI products.

• Suggested interim measures –

  • Pre‑wetted data sets (Rajani) – Clean, cleared data reduces compliance risk.
  • Democratized compute – Shared cloud/edge resources to lower cost barriers.
  • Lab‑to‑market pipelines – Universities should help translate research into market‑ready AI products.
  • Future‑skill development – Cross‑training of technologists in law/policy to bridge “code‑knows‑law” gaps.

3.2. Governance & Auditability (Anne)

• Beyond transparency – Emphasised auditability as a governance pillar to verify consent compliance and ownership tracking.
• Reference frameworks – Adoption of Singapore’s AI Verify and NIST AI Risk Management Framework as templates for Indian policy.

3.3. Licensing, Commons & Public‑Good Models (Sabeen)

• Hybrid licensing – Distinguish core public commons data (free) from sector‑specific, paid licences (e.g., health, finance).
• Funding model – Centralised pool (potentially government‑subsidised) to support universal data commons while charging for commercial exploitation.

3.4. Legislative Choice – Section 52 Exception (Shardul – final remarks)

• Proposed statutory amendment – Insert a TDM exception into Section 52 of the Copyright Act, enabling mass‑copying for model training without royalty.
• Balancing act – The amendment would nullify existing copyright barriers for LLM training while preserving downstream commercial rights.

3.5. Closing Vision – “Less is More” (Nitin)

• Innovation‑first mantra – Prioritise rapid AI development; treat IP regulation as surgical, not exhaustive.
• Targeted focus areas – Deep‑fakes, impersonation, enterprise‑level proprietary knowledge, consumer‑data privacy – as “high‑priority” problems.

---

4. Announcements & Closing

• NASSCOM announced a tree‑planting pledge – ten trees per speaker, with digital certificates to be issued.
• Group photograph to be taken after the session (as requested earlier).

Key Takeaways

• Tech‑legal synergy is essential – India should adopt a “techno‑legal” AI‑IP framework that blends open‑source acceleration with enforceable downstream IP rights.
• Clear ownership is the ROI catalyst – Without explicit IP ownership for training data and model outputs, AI projects cannot reliably demonstrate economic value (as highlighted by Chubb’s Rajani).
• Three‑tier data perspective – Policy must differentiate macro (public‑domain Gen‑AI), enterprise (proprietary), and individual (personal privacy) layers, each with distinct regulatory tools (Nitin’s model).
• Statutory text‑and‑data‑mining exception needed – A Section 52 amendment (proposed by Shardul) would unblock AI innovation while preserving creators’ future‑income rights.
• Enforcement technology is as important as law – Watermarking, provenance, and automated infringement detection (Rapid7’s Sabeen) are critical for practical IP protection.
• Open‑source foundations can be a competitive advantage – IBM’s stance shows that transparent foundation models, paired with downstream IP safeguards, accelerate Indian startup ecosystems.
• Creative‑sector protections – Opt‑out training flags and an explicit “impersonation right” (Adobe’s proposal) are needed to safeguard artists and performers.
• Institutional support for startups – Pre‑wetted datasets, shared compute resources, lab‑to‑market pipelines, and cross‑disciplinary skill development can bridge the legal‑technical gap.
• Governance must be auditable – Auditability mechanisms enable verification of consent and ownership, complementing transparency and explainability.
• Policy focus should be surgical, not sweeping – As Nitin emphasized, prioritize a few high‑impact AI‑IP issues (deep‑fakes, enterprise knowledge, privacy) and avoid over‑regulation that stalls innovation.

---

See Also:

• ai-for-everyone-empowering-people-businesses-and-society
• democratizing-ai-resources-in-india
• democratizing-ai-resources-and-building-inclusive-ai-solutions-for-india
• ai-for-economic-growth-and-social-good-ai-for-all-driving-economic-advancement-and-societal-well-being
• ai-and-indias-economic-growth-sectoral-impact-and-the-road-ahead
• ai-innovators-exchange-accelerating-innovation-through-startup-and-industry-synergy
• ai-for-inclusive-societal-development
• responsible-ai-at-scale-governance-integrity-and-cyber-readiness-for-a-changing-world