AI & Copyright
| Case | Ruling | Unresolved Issues |
|---|---|---|
| Thomson Reuters vs Ross AI | Training AI on copyrighted data = "Fair Use" if output is transformative. | Legality of pirated datasets. |
| Bartz vs Anthropic | AI training = "Transformative" (like human learning). | Compensation for copyright holders. |
| Kadrey vs Meta | No market harm proven → Fair Use applies. | Scope of "transformative" outputs. |
Net Effect: Tech firms gain legal cover for training, but piracy concerns & market impact remain contentious.
Global Legal Landscape
| Jurisdiction | Approach | Key Mechanism |
|---|---|---|
| USA | Flexible | Fair Use Doctrine (Case-by-case). |
| EU/UK | Conditional Permission | Text & Data Mining (TDM) Exceptions. |
| India | Evolving (No AI-specific laws) | Copyright Act, 1957 + Sec 52 (Fair Dealing). |
Core Conflict:
- Copyright Holders: Demand licensing/compensation for training data.
- AI Firms: Argue training is "transformative" and non-infringing.
Critical Copyright Challenges
- Replication Risk: AI outputs may mimic copyrighted works (e.g., art, text).
- Authorship Gap: Copyright requires human creativity → AI-generated works lack clear ownership.
- Data Sourcing: Scraping web data (including copyrighted material) without consent.
- Global Inconsistency: Patchwork of laws creates compliance hurdles.
India’s Position (ANI vs OpenAI Case)
Legal Framework
- Copyright Act, 1957: Grants exclusive rights (reproduction, adaptation) to creators.
- Section 52: Permits "fair dealing" for research/criticism (no AI-specific exemption).
- Enforcement: Civil/criminal remedies under IT Act + global treaties (Berne Convention, TRIPS).