On 1 July 2024 the 78th session of the UN General Assembly adopted by consensus a resolution on international cooperation for AI capacity building, introduced by China and co-sponsored by more than 140 states. Consensus means no vote against, which in turn means the text had to accommodate sharply different positions. Its weight lies less in enforcement than in agenda-setting: it is the first UN resolution devoted to AI capacity building.
Why capacity building
International discussion of AI governance has long centred on safety, ethics and standards, with capacity building placed in an annex. This resolution moves it to the centre and points to gaps in compute, data, talent and institutions across developing countries. If those gaps are not closed, rule-making remains an internal affair of a few states — the practical reason so many governments co-sponsored the text.
The term is broader than training. It covers institutional capability, access to evaluation tools and a voice in rule-making, alongside technical training and exchange. For a company entering an emerging market, these questions determine whether the local regulator can absorb the product at all.
The weight of the UN framework
The emphasis on the UN framework points at fragmentation. The OECD, the G7 and the G20 all produce AI principles, but their membership is limited. The General Assembly’s universality makes it one of the few venues where more than 140 states sit at one table.
Universality has a price. The text can only use language every delegation accepts, so fairness, inclusiveness and cooperation recur without executable metrics. Read as a treaty it disappoints; read as agenda-setting it is accurate. It fixes which issues will appear in bilateral dialogue, multilateral funds and technical assistance over the next few years.
Policy alignment for companies going abroad
When bidding, raising capital or responding to diligence abroad, Chinese AI companies are often asked which international frameworks they follow. Citing a General Assembly resolution is more stable than citing an institute’s principles: the text, the list of co-sponsors and the adoption procedure are public. That is verifiable evidence of alignment, not public messaging.
Granularity matters. A general claim of compliance carries little weight in professional diligence. Citing the specific paragraphs on technical cooperation and explaining how a project maps to them does. Where a project involves technology export, training or joint development, export-control rules and the host state’s foreign-investment rules apply in parallel.
Where capacity-building cooperation begins
The resolution creates no contracts, but it shapes the direction of multilateral agencies, development banks and technical assistance. Participants face the procurement rules of the funder, the admission and data rules of the host state, and the intellectual property arrangements of the project itself.
IP is often neglected. Who owns joint results, whether model weights can be deployed locally, whether training materials can be reused — all of this belongs in the proposal stage. Capacity building carries a public character, and hosts expect outputs to remain and be reusable. Commercial interests are protected through licence scope, service term and follow-on support, not by retaining every right.
For multi-jurisdiction projects and admission questions, see our multi-jurisdiction compliance service and business licensing service. Where data crosses borders, see data compliance. Definitions are collected in the glossary.
This article is based on the resolution on international cooperation for AI capacity building adopted by the 78th session of the UN General Assembly on 1 July 2024, and on public information. It is for reference only and does not constitute legal advice.