Artificial intelligence is becoming an increasingly common topic of conversation across archives, libraries and museums. From supporting cataloguing to identifying sensitive content and improving discoverability, AI offers new possibilities. But without careful preparation, it also carries risks.
The newly published AI Preparedness Guidelines for Archivists, developed through the FLAME (AI for Libraries, Archives and Museums) project and funded by the Archives and Records Association UK & Ireland, provide practical guidance to help the Galleries, Libraries, Archives and Museums (GLAM) sector approach AI adoption in a responsible and informed way.
We spoke with Professors Giovanni Colavizza and Lise Jaillant, who lead the FLAME project, about what it means for collections to be “AI-ready”, why metadata and documentation are central to responsible AI, and how this work connects to broader Open Science principles of transparency, accessibility and sustainability.
What prompted the development of the AI Preparedness Guidelines for Archivists? What gap did you see in the sector that needed to be addressed?
The guidelines respond to a growing pressure to apply AI in archives without sufficient consideration for archival principles, data quality, and ethical implications. While AI promises support for sensitivity review, metadata creation, and new access modes, we identify a gap in systematic guidance on how to prepare collections so that AI use remains grounded in provenance, authenticity, and awareness of bias. The sector lacked a practice-oriented framework linking data readiness, archival theory, and responsible AI deployment. This is the gap we aim to address.
The guidelines emphasise preparation before applying AI tools. What does it actually mean for a collection to be “AI-ready”?
“AI-ready” means systematically assessing, documenting, and preparing collections so that AI systems can operate transparently and responsibly. This includes documenting completeness and excluded data, improving item-level and contextual metadata, ensuring coherent file structures and formats while preserving provenance, and defining application-specific evaluation metrics.
AI readiness is not just technical standardisation; it is an interventionist and ethically informed approach to data curation that surfaces gaps, biases, and limitations.
Many organisations are under pressure to experiment with AI. What are the risks of applying AI to archival collections without proper preparation?
Without preparation, AI risks missing critical information, amplifying existing biases, or misrepresenting incomplete collections as comprehensive. Inadequate metadata, poor documentation of gaps, and inconsistent formats can also reduce accuracy and reliability, increasing hallucinations or misleading outputs in generative systems. More broadly, treating archives simply as data can flatten provenance, legal status, and social meaning, undermining archival integrity.
How do these guidelines support broader Open Science goals, such as accessibility, transparency, and long-term sustainability of research data?
The guidelines promote detailed documentation (e.g. datasheets, data-envelopes), clarity about completeness and exclusions, and alignment with FAIR principles for data access. By strengthening metadata, provenance tracking, and structured access, they enhance discoverability and interoperability. At the same time, by foregrounding documentation, auditability, and explicit evaluation metrics, they contribute to transparency and reproducibility in AI-supported research.
From a policy perspective, what would you recommend institutions or funders prioritise to support responsible AI adoption in the GLAM sector?
Institutions and funders should prioritise investment in metadata quality (including item-level and narrative metadata), coherent digitisation and file-structure practices, documentation of completeness and exclusions, and clear governance and evaluation frameworks. They should also support staff training, cross-disciplinary collaboration, and pilot projects that allow careful, evaluated experimentation before large-scale deployment.
What skills or organisational changes will be most important for archives and cultural heritage institutions in the coming years?
Key needs include developing staff literacy in AI capabilities and limitations, strengthening data curation and documentation practices, and fostering collaboration between archivists, technologists, and community stakeholders. Organisationally, institutions will need structured evaluation protocols, oversight mechanisms, and workflows that ensure AI outputs remain subject to human review and aligned with archival principles.
The guidelines are the first major public output of the FLAME project. What impact do you hope this work will have across the GLAM sector and beyond?
The work aims to provide a grounded, practical framework that enables archives to engage with AI on their own terms – treating AI as a supporting tool rather than a replacement for archival labour. By integrating data readiness, archival theory, and ethical governance, the guidelines seek to foster more responsible, transparent, and sustainable AI adoption in the GLAM sector.
As AI tools continue to enter research and cultural heritage workflows, initiatives like the FLAME project highlight a key message: responsible innovation begins with well-prepared, well-documented data. For research infrastructures, policy makers and institutions across Europe, investing in metadata quality, governance and skills development remains essential to ensuring that AI strengthens, rather than undermines, trust in our shared knowledge systems.
Check out the AI Preparedness Guidelines for Archivists at: https://www.archives.org.uk/ai-preparedness-guidelines-for-archivists
