context: The World AI Cooperation Organisation was formally established in Shanghai on 16 Jul 2026 when 29 founding member states signed its founding agreement. The organisation is the first intergovernmental body open to all sovereign states focused on inclusive AI development. It emerged alongside intensifying US export controls on advanced AI chips and model weights, which the PRC has characterised as attempts to maintain technological hegemony through supply chain fragmentation.
Global AI governance stands at a critical juncture, with two divergent paths: a PRC-led multilateral, open-source and development-inclusive approach anchored in WAICO (World AI Cooperation Organisation), and a US-led approach centred on export controls, exclusive small-multilateral mechanisms and technology bloc formation, according to a comparative analysis published by the Zhongjiao Open Source Innovation Research Institute.
The PRC model treats AI as a global digital public good and is built around four pillars, according to the institute
- sovereign equality and development priority
- the PRC's position submitted to the UN asserts that development rights are a fundamental human right in the digital era and that global AI rules must not be dictated by a small number of developed countries
- WAICO as a multilateral platform
- the organisation provides three core functions including rule coordination on large model risk classification and algorithm transparency standards, inclusive development through a dedicated fund for AI talent training and computing infrastructure in developing countries and joint risk governance covering deepfake detection and cross-border AI attack response
- non-discriminatory open-source sharing
- open-weight models including Qwen and DeepSeek are made available globally without geopolitical conditions, in contrast to what the institute characterises as the conditional and gatekept nature of US open-source releases
- balanced data governance
- the PRC's 2024 APEC data cross-border flow cooperation initiative allows countries to set their own data security boundaries while enabling commercial data to flow under multilateral recognition frameworks
The US approach is characterised by the institute as aiming to consolidate technological dominance through a combination of unilateral export controls, exclusive small-multilateral rule-setting via the G7 Hiroshima AI Process and the US–EU Trade and Technology Council and standards export. The July 2026 White House AI action plan 'Winning the Race' extends controls to frontier model weights trained beyond ten to the power of 26 compute operations, advanced GPUs and electronic design automation tools. While the US has open-weight models including Llama, the institute argues these are conditional instruments serving ecosystem lock-in rather than non-discriminatory sharing, with high-end weights remaining under strict export control.
The institute identifies five structural divergences between the two approaches
- multilateral versus unilateral governance
- non-discriminatory versus conditional open source
- open supply chains versus bloc fragmentation
- common comprehensive security versus exclusive absolute security
- digital civilisation pluralism versus single-standard export
These represent differences in underlying philosophy, global order vision and technology development objectives rather than merely regulatory style, the institute argues.
Six areas of constructive cooperation remain feasible, according to the institute
- joint governance of extreme high-risk AI including bioweapon-enabling models and autonomous lethal weapons
- cross-border digital crime enforcement coordination
- civilian technical standards harmonisation
- third-party AI capacity building in Africa, Southeast Asia and Latin America
- data security and privacy regulation mutual learning
- AI scientific research cooperation on climate and public health, where geopolitical competition is weaker and shared public value is high