On 18 October 2023 China released the Global AI Governance Initiative during the third Belt and Road Forum for International Cooperation. It is not a binding treaty, but it sets out a coherent set of governance propositions and helps explain China’s position in international AI rule discussions.
Its practical relevance is that, as AI rules in different jurisdictions compete, the initiative’s keywords will appear in regulatory engagement, industry standards and international cooperation projects.
Human-centred, AI for good
The initiative states that AI development should be human-centred, respect human dignity and rights and serve common wellbeing. This resembles the human-centred orientation of the EU AI Act, though the emphasis differs: the EU stresses fundamental rights and risk tiers, while the initiative stresses development direction and ethics.
For companies, this means AI value propositions cannot speak only of efficiency; they must explain effects on people. Scenarios involving minors, vulnerable groups, health and public safety require more careful design and assessment.
Sovereignty, fairness and non-discrimination
The initiative emphasises respect for national sovereignty, law and governance models, opposes using AI to interfere in other countries’ internal affairs, and opposes technological monopoly and bloc confrontation. It argues that developed and developing countries should have fair opportunities in technology access, capacity building and rule-making.
These propositions will shape international standards debates. Companies participating in standards, open-source communities and cross-border cooperation may encounter tensions between governance concepts. Compliance strategy should distinguish technical standards from governance propositions.
Risk tiers and agile governance
The initiative proposes tiered and classified AI risk management, agile governance and dynamically adjusted regulation. This echoes China’s tiered supervision of generative AI and the EU’s risk-based approach.
For companies, tiering means not applying one process to all AI functions. High-risk functions need stricter data, testing and human oversight; lower-risk functions can use lighter compliance paths. Product teams should maintain a function list and allocate compliance resources by risk.
A broadly participatory international mechanism
The initiative advocates a broadly participatory, consensus-based international governance mechanism, supports the role of the United Nations and opposes turning governance rules into tools of a few countries. The Bletchley Summit in November 2023 and UN discussions are practical settings for this proposition.
For companies, fragmented international mechanisms raise the cost of complying with multiple regimes. Companies expanding abroad need a “rules map”: which markets have dedicated AI legislation, which rely on existing data and consumer protection law, and which implement through standards and self-regulation.
How companies can use it
First, translate the principles into product review questions: are user rights respected, are outputs discriminatory, is the system explainable, can humans intervene. Second, map AI functions by risk tier and configure assessments and documentation accordingly. Third, follow international standards and self-regulation and participate where possible. Fourth, build a rules map for multi-jurisdiction operations rather than copying one jurisdiction’s solution.
For multi-jurisdiction AI compliance, see our multi-jurisdiction service; for the EU AI Act, see EU AI Act; for AI product launch documents, see AI product launch documents.
This article is based on the Global AI Governance Initiative released on 18 October 2023. It is for reference only and does not constitute legal advice.