The Report of the Independent Expert Group on Artificial Intelligence and Culture (CULTAI) was prepared for MONDIACULT 2025, the UNESCO World Conference on Cultural Policies and Sustainable Development, held in Barcelona from 29 September to 1 October 2025. UNESCO convened ten international experts and two youth representatives, chaired by Salim Dada, to examine how AI is changing the ways culture is created, distributed, and preserved, and to propose policy responses. The experts participated in their personal capacity; the findings are their own and do not represent UNESCO’s official position.
The report starts from three converging problems: AI is accelerating faster than cultural governance can adapt, digital divides persist between regions and communities, and ethical risks are escalating, from deepfakes to systemic bias. It organizes its analysis into seven thematic sections:
- The ethical landscape of AI in culture;
- Cultural rights in the digital era;
- AI and cultural value chains in the creative economy;
- Cultural education in the AI era;
- Climate, culture, and environmental sustainability;
- Endangered heritage and AI; and
- AI, freedom, and cultural sovereignty.
Across these sections, the report proposes measures that include transparency about training data, opt-in collective licensing and compensation funds for creators, diversity requirements for recommendation systems, labelling of synthetic content, public AI infrastructure and open-source alternatives, community-governed language models, and shared international standards for the digital documentation of heritage. Its most distinctive contribution is conceptual. It distinguishes explicit cultural expressions, which intellectual property law partly protects, from implicit and latent ones: the collective patterns that emerge when large volumes of everyday digital data are aggregated to train AI. The report argues that this second category should be treated as a collective common good, governed through three pillars: recognition, access, and benefit-sharing.
As a policy document, the report treats AI as a question of cultural rights and sovereignty rather than only of innovation or market competition. It gives governments, cultural institutions, and communities a framework for asking who controls cultural data, who benefits from its use, and who sets the terms.
TRANSFERABLE LESSONS
1. Treat cultural data as a shared resource with collective rights.
Cultural data includes individual works and the aggregated patterns of community life that AI systems learn from. Policy should recognize that collective dimension and give the communities that produce the data a role in deciding how it is used.
2. Put consent and terms of use where the data travels.
The report recommends embedding creator consent and conditions of use directly in the metadata of digital cultural goods, and building open repositories with clearly defined usage rights. Terms that move with the data protect it beyond any single platform.
3. Build community-governed infrastructure and keep revising the rules.
Examples such as Te Hiku Media in Aotearoa New Zealand and Tirtha in India show communities keeping control of their data and tools. Because AI changes quickly, governance works best as an ongoing process of testing and revision rather than a fixed document.