Indo-US Working Group on Technology Policy: AI Training Data and Copyright

This session of the Indo-US Working Group on Technology Policy focused on India's proposed framework for AI training data and copyright.

India is weighing new rules that would require AI developers to pay royalties when they train models on copyrighted material. In December 2025, India's Department for Promotion of Industry and Internal Trade (DPIIT) published a working paper proposing a mandatory blanket license for AI developers to use copyright-protected material for training, with royalties paid to a centralized collection body. This session examined the proposal, with participants from both countries sharing perspectives on its workability.

On the economic side, participants raised concerns about royalties calculated as a share of global revenue. If multiple jurisdictions adopt similar requirements, those obligations could stack beyond what any company earns in total. Participants suggested that any viable framework would need to tie obligations to local revenue rather than global earnings. They also noted that the proposal leaves key definitions open: the point at which a royalty obligation is triggered, whether at the deployment of a commercial application programming interface (API), a free product with a paid tier, or a research model later adapted for commercial use, remains unclear.

Participants also raised legal concerns. Several noted that the framework rests on a theory that AI outputs will substitute for original works, but that no demonstrable harm to rightsholders had been established before the proposal came forward. They flagged this as a potential vulnerability under Indian constitutional protections for the right to trade and free expression. (Shortly after the session, the Delhi High Court refused ANI Media's application for an interim injunction against OpenAI on July 24, 2026, holding on a prima facie basis that training a large language model on ANI's news content can fall within India's statutory fair-dealing exception.)

Participants also raised questions about who would actually benefit from a mandatory licensing system. Several noted that Indian copyright enforcement has historically favored intermediaries over original creators, and expressed concern that a new collection mechanism layered on top of existing structures could reproduce that pattern at scale.

The session also took up the question of where regulation should focus. Some participants argued that training-data transparency is technically difficult to achieve in practice, since AI training reduces source material to mathematical weights rather than stored copies, and suggested that output-focused regulation may be more tractable. In regards to training on pirated material, participants noted agreement: no one at the session defended the practice, and several confirmed their organizations use only lawfully accessed content.

On the regulatory timeline, participants noted that India's IT secretary had stated that existing domestic laws were sufficient to address AI-related concerns, and that the IT Ministry rather than DPIIT is likely to hold final authority over any framework. Participants observed that India amended its IT Rules in February 2026 to introduce the country's first binding AI-specific compliance obligations, and noted that the pace of legislative activity makes the timeline for further action difficult to predict.

The Indo-US Working Group on Technology Policy is a Track II dialogue convened by The Dialogue and the Information Technology and Innovation Foundation (ITIF). Session summaries do not attribute comments to individual participants.