After recommending a compulsory blanket-licensing regime for training AI systems in its first working paper, the government panel is set to release a second working paper on authorship and copyrightability of AI-generated content, in the next two months, a senior government official said on Thursday.
Addressing AI Authorship
The official said the committee’s work on the forthcoming paper will analyse the copyrightability of machine-generated works, the boundaries of authorship and the extent to which human intervention, such as prompting can be treated as creative labour.
This is described as a natural progression from the first working paper, which dealt exclusively with the use of copyrighted content in AI training.
The group spearheading this was constituted on April 28 this year by the Department for Promotion of Industry and Internal Trade (DPIIT). The eight-member panel is headed by Additional Secretary Himani Pande and includes lawyers, academics and industry representatives.
“The same committee will continue to work on the second working paper which will be on copyrightability of the AI generated works and their operations,” Pande said while speaking to reporters.
Retrospective Royalty Framework
The first paper, unveiled earlier this week, proposed a “One Nation, One Licence, One Payment” model that mandates a blanket licence for AI developers over all lawfully accessed copyrighted works. The committee argued that creators have been left uncompensated even as large AI models routinely ingest their work, while AI firms insist that any restriction on training data risks undermining model accuracy and amplifying bias.
To resolve these competing pressures, the working paper extends blanket access to AI companies but requires them to share a portion of future revenues with creators. Currently, the official said that the Ministry of Electronics & IT (MeitY) is also in board and involved in the consultation process.
A key feature of the proposal is its retrospective application to payment obligations. Even if a model was trained before the legal framework comes into force, the AI developer would still be liable to pay royalties once the tool is commercialised. The official justified this approach by noting that once the revenue and the economic benefit is derived from the copyrighted content after deployment, the possibility is that it is largely trained on previous datasets. By contrast, the lawful-access requirement will not apply retrospectively.
Government officials said that the system relies on coarse-level data disclosures rather than granular dataset reporting. AI firms would have to reveal only the proportion of literary, artistic, audio visual or other categories of content used in training. If a creator disputes an AI company’s declaration, the burden of proof shifts to the developer.
The committee believes that these reforms cannot be implemented immediately. Any operationalisation will require amendments to the Copyright Act, followed by the creation of Copyright Remuneration Collective for AI Training (CRCAT) and the sector-wise collective management structures. Members cautioned that this is a long process, alongside describing the framework as a necessary step to remain competitive in the global AI race.
