This is the final report of the UN Secretary-General’s High-Level Advisory Body on AI, the document that framed AI governance as a global, not national, problem. It concludes that no existing framework is truly global in reach, and names three gaps, representation, coordination and implementation, that leave most of the world outside the conversation.
The headline
The body argues that AI’s raw materials and applications are borderless, so its governance cannot be left to markets or to a handful of States. It documents a patchwork of hundreds of principles and frameworks that, taken together, still cover no one comprehensively. Its answer is a light, networked architecture of seven recommendations: an international scientific panel on AI, a twice-yearly policy dialogue on AI governance, an AI standards exchange, a capacity development network, a global fund for AI, a global AI data framework, and a small AI office in the UN Secretariat to connect them. Two of these, the scientific panel and the policy dialogue, were subsequently established by the General Assembly.
The numbers
- 7 recommendations. The core of the report, from a scientific panel to a UN AI office (recommendations 1–7, pp. 10–20).
- 7 versus 118. Across seven prominent non-UN AI governance initiatives, seven countries are parties to all of them, while 118 countries are parties to none, primarily in the global South (p. 8).
- Whole regions absent. Of countries not involved in any sampled initiative: 48 of 54 in the African Group, and 44 of 54 in the Asia and the Pacific Group (p. 8).
- Zero of 100. Of the top 100 high-performance computing clusters capable of training large AI models, not one is hosted in a developing country (p. 14).
- 10% of grants. From 2018 to 2023, only 10 per cent of grants for AI initiatives addressing the SDGs went to organizations in low- or middle-income countries; for private capital the figure was 25 per cent, over 90 per cent of it in China (p. 27).
- 348 experts, 68 countries. In the body’s risk pulse check, 7 in 10 experts polled were concerned or very concerned that AI harms would grow substantially more serious in the next 18 months (p. 28).
The counting angle
The report’s sharpest contribution is a headcount of who is in the room. Its representation gap is a literal tally: seven countries at every table, 118 at none. That inequality of voice is mirrored by an inequality of inputs, no top compute cluster in a developing country, and by what the body calls “missing data,” the large parts of the globe that are data poor and therefore invisible to the models trained on the rest. Its proposed global AI data framework and capacity network are, at root, attempts to make the uncounted countable, so their languages, contexts and needs register in the systems being built.
Watch & read
- Governing AI for Humanity, the full final report (United Nations High-Level Advisory Body on AI).
- Independent International Scientific Panel on AI, the standing panel this report proposed and the General Assembly created.
Figures are drawn from the report as published in September 2024. The representation figures cover a sample of seven non-UN initiatives and exclude the UNESCO ethics recommendation and the two 2024 General Assembly resolutions on AI.