⚡ Quick Summary
What Should be Internationalised in AI Governance is a research paper on which AI policy issues genuinely warrant international coordination. It argues that the field lacks an agreed basis for setting priorities, so international activity can be unfocused, duplicate work or fail to achieve its goals, and it proposes a framework for judging individual issues. The central device separates the demand for internationalisation — the extent to which an issue produces cross-border effects that no single state can manage — from the supply of internationalisation, meaning the feasibility of reaching and sustaining agreement. It also uses the legalisation vocabulary of principles, standards and rules to describe the forms internationalisation can take.
The framework is applied to concrete issue areas, including data governance (data protection, data sovereignty and data localisation), compute governance and semiconductor supply chains, and frontier AI model governance such as evaluation and risk mitigation. The paper ends with conclusions for policymakers and is supported by appendices and an extensive bibliography.
🧩 What’s Covered
The document is organised as a paper with numbered sections, appendices and a long reference list.
- Introduction: frames the coordination problem, notes that states increasingly recognise the need to align AI policy while it remains unclear which policy issues should be prioritised in international discussions, and sets out the risk that unclear priorities waste effort and resources.
- Internationalisation: defines the concept and builds the analytical framework, distinguishing the demand for internationalisation from its supply and describing the institutional forms — principles, standards and rules — that an international arrangement can take.
- Analysis of specific issues: applies the framework issue by issue, in clusters that include data governance (data protection, data governance, data sovereignty and data localisation), compute governance and AI chips, and frontier AI model governance such as evaluation and risk mitigation.
- Assessment and conclusions: brings the demand and supply judgements together into priorities for policymakers and states which issues merit international treatment.
- Appendices: provide supporting definitions and taxonomies of the issues and governance functions discussed.
- References: an extensive bibliography running through the final pages of the document (roughly pages 61–75).
💡 Why it matters?
For anyone deciding where scarce governance capacity should go, the paper offers a test rather than a wish list: it asks whether an issue has cross-border effects that single states cannot manage and whether agreement is feasible given interests, incentives and institutional capacity. That is useful for policy teams preparing or assessing international initiatives, for conformity and standards work that must align with whatever is agreed internationally, and for organisations tracking compute and frontier-model governance, where international arrangements are still developing.
❓ What’s Missing
As far as the extracted text shows, the paper stays at the level of framework and priorities. It sets out no measurable rubric for scoring demand or supply, so different analysts may reach different conclusions on the same issue, and it does not quantify how much coordination a problem needs or what it would cost. Institutional design is largely left open: membership, decision rules, enforcement, funding and the reform of existing bodies are not worked through, and there is no sequencing or timetable for negotiation. No version label or date was readable in the extraction.
👥 Best For
Policy advisers and analysts deciding what to put on an international AI agenda; officials preparing positions for summits, standards bodies or export-control forums; researchers mapping the international governance landscape; and governance teams weighing whether a domestic rule should follow an international baseline.
📄 Source Details
The extraction identifies the document as What Should be Internationalised in AI Governance, a 75-page English-language paper with numbered sections, appendices and a long bibliography; no publisher, author list, date or version label could be read reliably. The PDF's text layer is badly encoded — most letters appear as unusable glyphs — so the front matter, quotations and headings in this review should be checked against the original file.
