AI4LAM, National Library of Norway, Stanford University

A successful Global Dialogue on AI Governance would, from perspective of cultural heritage, be one that brings together diverse voices across sectors—cultural heritage institutions, academia, policymakers, and technology communities—and creates a shared understanding of both the opportunities and risks that AI presents for memory organizations. A key outcome would be the recognition of cultural heritage data as a critical component of the global AI ecosystem. This includes acknowledging the value of libraries, archives, and museums as stewards of high-quality, trusted, and context-rich data, and ensuring their role is reflected in emerging governance frameworks. Equally important would be the establishment of practical, actionable principles for the responsible use of AI in and from this domain. This includes guidance on transparency, data provenance, ethical reuse, and respect for copyright and privacy. A successful dialogue would move beyond high-level statements toward identifying concrete areas for collaboration, such as shared standards, tools, and best practices. We would also see success in the creation of lasting connections and commitments—mechanisms for ongoing international cooperation that include governance and management of (digitized, digitally born, and harvested) data. Commitments should support knowledge exchange, capacity building, and inclusive participation, particularly for institutions with limited resources. Finally, the dialogue should help articulate a balanced approach to AI governance—one that safeguards rights and cultural integrity while enabling innovation and access. Setting a clear direction for future work, including contributions to policy development and collaborative outputs such as white papers, recommendations or guidelines, would mark a strong and meaningful outcome.

• Safe, secure and trustworthy AI
• AI capacity-building
• Social, economic, ethical, cultural, linguistic and technical implications of AI
• Open-source software, open data and open AI models

AI4LAM connects the global community of libraries, archives, and museums with key priorities in contemporary AI governance and practice. At its core, AI4LAM promotes the development and use of safe, secure, and trustworthy AI by advocating for transparency, accountability, and responsibility, as well as implementation in cultural heritage sector. Cultural heritage institutions play a crucial role, as they curate high-quality, well-documented, and context-rich data that can support more reliable AI systems. A central focus of AI4LAM is AI capacity-building. By fostering knowledge exchange, training, and collaborative experimentation, the community helps institutions of varying sizes and resources develop the skills needed to engage critically and effectively with AI technologies. This includes not only technical competencies but also policy awareness and ethical literacy. AI4LAM is also deeply engaged with the social, economic, ethical, cultural, linguistic, and technical implications of AI. Its members work to ensure that AI systems respect cultural diversity, multilingualism, and the integrity of heritage collections, while addressing issues such as bias, representation, and equitable access. In doing so, AI4LAM contributes to shaping AI that reflects a broader range of human knowledge and experience. Finally, AI4LAM strongly supports open-source software, open data, and open AI models as enablers of transparency, collaboration, and innovation. Through these interconnected areas, AI4LAM helps position cultural heritage institutions as active contributors to a more inclusive, responsible, and sustainable AI ecosystem.

Cross-sectoral collaboration: a critical emerging issue. Effective AI governance increasingly depends on sustained cooperation between cultural heritage institutions, academia, public authorities, civil society, and the private sector. However, these sectors often operate with different priorities, vocabularies, and regulatory constraints. Strengthening structured collaboration mechanisms is essential to ensure that governance frameworks are both practical and inclusive. Sustainability: both environmental and institutional is an emerging concern. AI systems require significant computational resources, raising issues of energy consumption and environmental impact, as well as the long-term financial and infrastructural sustainability for institutions adopting AI. Infrastructure dependency and development of parallel systems: for ex. science (EOSC) / culture (ECCCH). Infrastructures deserve more explicit recognition, building autonomy, connecting already existing systems, such as AI Factories or data centers use for multiple purposes, shape access conditions, and influence governance practices in ways that are not always transparent or equitable. Preservation of AI systems and outputs: an underexplored area. Beyond using AI, there is a growing need to preserve AI-training datasets, generated content, models, and decision-making processes as part of the cultural and scientific record. Evaluation and measurement frameworks: still developing. There is a need for shared methodologies to assess the impact, risks, and benefits of AI systems in different contexts, particularly in the cultural heritage domain.

One of the most pressing challenges is the lack of harmonised governance frameworks for the use of AI in cultural heritage data. Institutions are increasingly approached by big tech, but also experimenting with machine learning for discovery, description, and access on their own, yet they often do so in a fragmented legal and ethical landscape. Uncertainty around copyright, privacy, and data protection—particularly under frameworks such as the GDPR—can slow down innovation or lead to overly cautious approaches that limit experimentation. At the same time, there are clear capacity gaps. Many libraries, archives, and museums lack the technical infrastructure and human expertise needed to responsibly deploy and evaluate AI systems. This creates uneven adoption across regions and institutions, reinforcing existing digital inequalities. Dependence on external platforms and proprietary tools further increases vulnerability and reduces institutional control over data and workflows. Despite these challenges, there are significant opportunities emerging through collaborative networks such as AI4LAM and events like Fantastic Futures. These initiatives enable knowledge exchange, co-development of tools, and the articulation of shared principles for responsible AI use. They also support the development of open datasets, open-source tools, and experimental environments. In particular, work around digital cultural heritage, web archives and harvested data is opening new possibilities for large-scale cultural and historical analysis, while also highlighting the need for better standards for provenance, transparency, and sustainability. Overall, while governance gaps currently create uncertainty and fragmentation, they also create space for innovation. The sector is increasingly positioned to shape more inclusive, trustworthy, and sustainable AI practices through cross-institutional and cross-border collaboration.

The AI Dialogue can play a crucial role as a bridge between global policy discussions and sector-specific practice. There remains a persistent gap in translating high-level principles into operational, context-sensitive guidance. The AI Dialogue is uniquely positioned to address this gap by fostering structured engagement between policymakers, technical experts, and practitioners working with real-world datasets and systems. Institutions, such as libraries, archives, and museums, manage long-term, high-value datasets that raise specific challenges around provenance, copyright, privacy, bias, and long-term preservation. Ensuring that perspectives of cultural heritage institutions are systematically integrated into global governance discussions is essential for producing frameworks that are both practical and legitimate. The AI Dialogue can this way elevate the voice of cultural heritage communities within international AI governance debates, ensuring that their role as custodians of trusted knowledge infrastructures is more fully recognised. This includes acknowledging their contribution not only as data providers, but also as key actors in AI infrastructural systems that serve public value. By fostering cross-sectoral and cross-border cooperation, the AI Dialogue can further support the alignment of AI standards while respecting regional and institutional diversity.

The AI Dialogue should build upon a growing ecosystem of initiatives that already operate at the intersection of AI governance, cultural heritage, and open knowledge infrastructures. Within the cultural heritage and memory institution domain, networks such as AI4LAM, IIPC, IIIF, Data Space for Cultural Heritage, AI4DH, AI Factories (for ex. PIAST-AI, PHAROS), IFLA, CENL, Operas, UNESCO IRCAI, and global events like Fantastic Futures provide important foundations for experimentation, knowledge exchange, and the development of shared practices around AI, data stewardship, and digital preservation. These communities have been instrumental in advancing applied work on AI in libraries, archives, and museums, particularly in relation to data-rich environments such as digitised heritage, web archives and born-digital collections.

To ensure meaningful participation, the AI Dialogue should adopt a structured, multi-layered format. A combination of plenary sessions and focused thematic working groups would allow for both high-level alignment and deep technical or sectoral exchange. Thematic tracks could include areas such as data governance and provenance, open AI ecosystems, capacity-building, and interoperability of governance frameworks. Each track should be co-led by stakeholders from different sectors to ensure balance and shared ownership.

First and foremost, the cultural heritage sector remains underrepresented in global AI governance discussions, despite its critical role as a steward of trusted, high-quality, and historically significant data. Libraries, archives, and museums manage complex, multilingual, and sensitive collections that are increasingly used for AI training, analysis, and knowledge extraction. However, their perspectives are often not systematically integrated into global policy debates, particularly regarding data governance, provenance, long-term preservation, and ethical reuse of cultural and historical materials.

The most effective engagement format for the AI Dialogue would prioritise cross-sectoral exchange as its core principle. This means bringing together participants from diverse backgrounds, regions, sectors, and levels of governance who are directly involved in the management, development, and use of data and AI models. Such a format should be designed as a structured but open exchange space where policymakers, technical experts, researchers, civil society actors, and practitioners from domains such as cultural heritage, science, tech, including initiatives like AI4LAM, can engage on equal footing. The emphasis should be on real-world experience with data governance, AI deployment, and model development, ensuring that discussions are grounded in practice rather than remaining purely theoretical. The format should enable continuous interaction rather than one-off consultation. This could include recurring thematic dialogues and collaborative working sessions. Hybrid participation models and multilingual support would further ensure inclusivity and global reach.

For ex.: Frameworks such as UNESCO's Recommendation on the Ethics of Artificial Intelligence and the OECD AI Principles provide widely recognised normative foundations for trustworthy and human-centred AI. Then European Comision AI Factories strategic priority, the Apply AI Strategy, the AI in Science strategy, the Resource for AI Science in Europe (RAISE), European Data Union Strategy etc. In terms of practical governance models, the Norwegian model of authors' reimbursement for text and data use in digital environments offers an important example of a rights-based and negotiated approach to data access. In this model, authors are compensated through collective agreements when their works are used, including in digital and increasingly AI-related contexts. This provides a concrete mechanism for balancing innovation with fair remuneration and respect for intellectual property, offering a potential reference point for discussions on training data used in AI systems.