Can We Build a Copyleft License for Community Data in the Age of AI

Key Focus

As community-generated datasets become increasingly important for training AI systems, there is a growing need to explore whether existing open data licenses adequately protect community contributions. This session will examine whether copyleft principles, which have successfully supported open source software ecosystems, can be adapted to datasets used for AI development. The discussion will focus on attribution, reciprocity, data provenance, community ownership, and the governance of data commons.

Brief

Community-led initiatives across India and the world are creating datasets that support language technologies, accessibility tools, and AI research. However, many of these datasets are released without licensing frameworks specifically designed for AI use. This creates a risk that community-generated data may be incorporated into proprietary AI systems without attribution, transparency, or any meaningful contribution back to the communities that created them. Open source software have relied on copyleft licenses such as the GNU GPL to ensure that software and improvements remain available to the community. As community-generated datasets become increasingly important for training AI systems, a similar question is emerging for data i.e. can copyleft principles and licence be applied to datasets used for machine learning?

A draft Community Data License has been developed by Viswam.ai and SFLC.in to address these concerns by applying copyleft principles to datasets. The proposed license seeks to ensure that community-generated datasets remain part of a shared commons by requiring attribution to original contributors, transparency regarding data provenance, and the sharing of modified or derived datasets under the same licence terms.

This session will use the draft license as a starting point for a broader discussion on the future of community data governance, open AI, and digital commons. The objective is to gather perspectives from developers, researchers, open-source contributors, civil society organisations, policymakers, and dataset creators on whether new licensing and governance approaches are needed to protect community data commons and how such a license could work in practice.

Discussion Points

1. Are existing open data and open-content licenses sufficient for AI training datasets?
2. How can community data commons be protected from extraction?
3. Can copyleft principles be meaningfully applied to data used for machine learning?
4. What should qualify as a derivative work in the context of AI systems?
5. Should AI developers have obligations towards the communities that create datasets?
6. How should attribution work when datasets are incorporated into AI models?
7. Is open-source AI meaningful without transparency regarding training data and provenance? Are model weights derived from training data?
8. What mechanisms can support reciprocity and contribution back to community-created data commons?
9. How can compliance with community data licenses be verified and enforced?
10. What challenges and opportunities arise when balancing openness, innovation, and community rights?

Discussion Format

Open discussion with participants. The session will begin with a brief introduction to the problem statement and the draft Community Data License, followed by a moderated discussion. Participants will be encouraged to share experiences, concerns, use cases, and perspectives from technical, legal, policy, research, and community viewpoints. The moderator will guide the discussion around key questions and explore broad conversation focused on the topic.