AI and Copyright Laws:

22 Jul 2025 GS 2 Governance

Click to view full image

The central legal controversy is whether training generative AI on copyrighted content and its resulting outputs infringe copyright laws.

AI models are trained using large datasets, often scraped from the internet, which include both public domain and copyrighted content.
The legality depends on whether AI outputs are transformative (adding value) or substitutes (harming the original market).
There's global legal ambiguity—some countries permit exceptions under fair use (US), text/data mining (EU, UK), or temporary copying provisions.

Anthropic Case:
- Judge: William Alsup.
- Ruling: Training on copyrighted material may qualify as transformative fair use.
- Caveat: Using pirated content may still lead to liability.
Meta Case:
- Judge: Vince Chhabria.
- Ruling: Use of copyrighted works by Meta didn’t cause market harm; thus, fair use applied.
- Observation: Tech firms should share profits with copyright holders.
Conclusion: Courts favour broad fair use, but highlight concerns about piracy and lack of creator compensation.

Legal uncertainty persists regarding:
- Whether training on IP-protected data is infringement.
- Ownership of AI-generated content.
- Global harmonisation of IP laws for AI is lacking.
Many IP laws were created before AI existed, creating gaps in regulation.

ANI vs. OpenAI case may provide clarity.
Under Copyright Act, 1957, using copyrighted work for reproduction, adaptation, etc., needs permission unless covered under Section 52 (fair dealing).
India’s stand: Existing IP laws are adequate for AI governance.
India allows enforcement through civil and criminal remedies, and supports international IP treaties.