DPIIT signals Copyright Act changes to address AI issues
Context
- The Government of India is planning major amendments to the Copyright Act, 1957 within the next three years to address challenges posed by Artificial Intelligence, especially Large Language Models (LLMs).
- DPIIT (Department for Promotion of Industry and Internal Trade) has released a working paper proposing a new licensing and compensation framework for online content used for AI training.
Why AI Is a Copyright Challenge?
- LLMs like ChatGPT train on massive datasets scraped from public websites.
- Publishers argue their copyrighted content is used without consent or compensation.
- AI companies argue that:
- Scraping publicly available data is fair use, and
- Training does not store or reproduce content directly.
DPIIT Proposal: “Blanket Licensing” Framework
Key Features
- Allow AI firms to scrape online content for training LLMs.
- Introduce a mandatory blanket license requiring AI firms to pay royalties once their models are commercialised.
- Royalties will be collected and distributed by a copyright society named:Copyright Royalties Collective for AI Training (CRCAT).
- Royalty distribution will be based on the contribution of websites whose data was used.
- This seeks to balance —
- Innovation and AI growth, and
- Compensation to content creators.
Rationale
- Similar tensions are seen globally (e.g., NYT vs. OpenAI lawsuit).
- India wants a middle-ground, unlike the U.S. (permissive) or EU (strict opt-out rules).
Additional Issues Under Study
DPIIT will release a second paper addressing:
- Whether AI-generated works can be copyrighted
- Who is the author: AI? Developer? User?
- Liability for infringementThis will inform the eventual amendments to the Copyright Act.
Opposition / Concerns
A. Big Tech & Industry (Nasscom)
- Nasscom (representing Google, Meta, Amazon, etc.) dissented.
- Key objections:
- Publishers should have an opt-out mechanism.
- Mandatory blanket licensing creates legal uncertainty.
- Could lead to more disputes, not fewer.
B. Burden of Proof Reversal
- Current copyright law:Copyright owner must prove infringement.
- DPIIT proposal:AI developer must prove they did not use someone’s content.
- Firms argue this is:
- Technically infeasible → AI models are probabilistic, not deterministic
- Legally inconsistent with copyright jurisprudence
- Creates unmanageable compliance burdens
Global Context
- EU AI Act: Requires transparency, data documentation, opt-outs.
- U.S.: Fair use often favours tech companies; lawsuits still ongoing.
- Australia, Canada: Moving towards royalty-based systems for news publishers.
India’s approach attempts a hybrid system balancing creators and AI firms.
Implications of the DPIIT Proposal
Positive
- Provides a legal pathway to train AI on large datasets.
- Ensures compensation to publishers and creators.
- Reduces litigation risks by creating an official licensing structure.
- Encourages development of domestic AI models.
Challenges
- Identifying data contribution percentages for royalty distribution.
- Determining authorship for AI-generated works.
- Potential increase in compliance costs for startups.
- Risk of creating a monopolistic or bureaucratic copyright society.
- Fear of overregulation stifling AI innovation.
Way Forward
- Establish transparent opt-out / opt-in mechanisms.
- Maintain traditional copyright principle: burden on claimant.
- Build clarity on fair use and text & data mining (TDM) exceptions.
- Ensure data usage respects privacy and national security.
- Create an independent, well-governed CRCAT to avoid monopolisation.
- Promote open datasets for Indian AI innovation.
