DPDP Rules status: compliance window active. Read the update

Skip to content
AI Policy · Copyright

AI and copyright: the DPIIT working paper

A DPIIT committee wants AI firms to get a licence to train on copyright works. Creators would get paid. This is a proposal. It is not law yet.

Educational only · not legal adviceReviewed by Sandesh Kokate, Editor

Short answer

In December 2025, DPIIT published Part 1 of a working paper on generative AI and copyright. A committee wrote it. The paper asks for feedback from the public. Nothing in it binds you today.

What it means for you

If you train AI models in India, watch this paper. Under the plan, you could train on any work you got lawfully. You would not need to ask each owner. But you would have to pay royalties.

Good practice: keep a record of where your training data came from and how you got it. The proposal only covers works that were lawfully accessed.

How it works

  • A blanket licence. The paper calls it a "hybrid model" (section 5.1). AI developers get a mandatory blanket licence to use all lawfully accessed works for training.
  • No opt-out. Owners could not keep their works out of AI training (section 5.1).
  • Pay for use. Owners get a legal right to payment. A share of revenue from AI systems trained on such works would be paid as royalties.
  • One collecting body. A nonprofit, the Copyright Royalties Collective for AI Training (CRCAT), would collect the money (section 5.3). The government would designate it.
  • Not everyone agreed. The paper records that Nasscom disagreed in its 17 August 2025 submission. Nasscom wanted a text and data mining exception instead.

What's still open

  • PIB's release of 9 December 2025 says DPIIT opened the paper for comments for 30 days. That window has closed.
  • Part 2 is pending. It will cover who owns AI outputs and who is liable when outputs infringe.
  • Any change would need an amendment to the Copyright Act, 1957. No bill has been confirmed by us.

Related: the AI Governance Guidelines, 2025.

Official text — Working paper, section 5.1
A Hybrid Model is proposed wherein a mandatory blanket license with a statutory remuneration right for the creators and copyright holders would be established for the use of all lawfully accessed copyright-protected works in the training of AI Systems. The rightsholders will not have the option to withhold their works for use in the training of AI Systems.
Source
DPIIT, Working Paper on Generative AI and Copyright, Part 1 (December 2025)

opened DPIIT working paper PDF

Advisory — not binding — checked 2026-10-06. Opened the DPIIT Working Paper on Generative AI and Copyright, Part 1.

Source
PIB release 2200741, 9 December 2025

opened PIB release

Advisory — not binding — checked 2026-10-05

← Back to AI Policy hub