Beijing is seeking to turn years of domestic regulation into practical tools for global AI governance. Whether other countries recognise its certificates, labels and testing systems will determine how far that experience can spread.
SHANGHAI — China is seeking to translate its domestic approach to artificial intelligence regulation into technical standards and governance tools that other countries can use, as governments move beyond broad ethical principles towards rules that can be implemented and enforced.
The push was on display at the World Artificial Intelligence Conference in Shanghai in July, days after the United Nations held its first Global Dialogue on AI Governance in Geneva.
Twenty-nine countries signed an agreement in Shanghai to establish the World AI Cooperation Organization, or WAICO. The conference also produced an industrial AI governance framework developed through a United Nations initiative with Chinese and British institutions, proposals on trust and interoperability between AI agents, and an international action plan on AI ethics.
China also pledged training and technology cooperation for developing economies.
The announcements raise a larger question: can the regulatory experience China has built at home become useful abroad?
China has introduced rules covering recommendation algorithms, generative AI services and the labelling of synthetic content. It has also begun linking AI governance to industrial testing, certification and the environmental impact of data centres.
But exporting that experience will require more than publishing regulations. Foreign governments and companies would need to recognise Chinese-backed certificates, read the same digital watermarks and trust test results produced under different regulatory systems.
“Some of those principles have pretty much settled down,” John Higgins, chair of the International AI Governance Association, said on the sidelines of the Shanghai conference. “There’s a convergence around common principles pretty much everywhere in the world.”
“Now we’re seeing the work to build on those principles.”
For much of the past decade, international discussions focused on whether AI should be fair, transparent, safe and subject to human oversight. The next stage is more technical: labels that users can see, certificates that auditors can issue and test reports that regulators are prepared to accept.
Turning principles into practical instruments is “not easy, but it can be done,” Higgins said. International standards bodies could play an important role because they bring together engineers and companies from different countries to work on common technical problems.
Speakers at an AI governance forum held alongside the conference highlighted four areas in which China’s large domestic market has produced tools that could potentially be used more widely.
The first is certification.
Huawei received ISO/IEC 42001 certification in March across five core business groups and nearly 100 sites. The standard sets requirements for how organisations manage AI systems throughout their life cycles. The certification was audited and issued by Swiss testing and certification company SGS.
Gan Bin, Huawei’s vice-president for standardisation and industry development, said the company wanted AI governance requirements to be internationally aligned so that developers would have clear rules when designing products.
The second area is the labelling of AI-generated content.
Chinese rules that took effect on September 1, 2025, require synthetic text, images, audio and video to carry visible notices as well as machine-readable information embedded in the files.
The metadata is intended to allow platforms to identify AI-generated material even after it has been copied or uploaded elsewhere. A national standard defines how the labels should be applied.
Such a system would have greater international value if platforms and regulators outside China could recognise the same metadata. Without interoperability, a watermark may stop working as soon as the content crosses a digital border.
The third area is sector-specific testing.
In the automotive industry, AI is increasingly used in manufacturing, software updates and driving decisions. Fan Hailong of the China Merchants Testing Vehicle Technology Research Institute said compliance should be built into the development process, from factory testing to over-the-air updates and maintenance records.
Those records could form an audit trail that helps establish responsibility after a crash or system failure.
An intelligent vehicle, Fan said, is “a moving AI terminal.”
The fourth area is the environmental cost of AI.
Zhang Jian, deputy dean of Tsinghua University’s Institute of Climate Change and Sustainable Development, said sustainability had been included among WAICO’s objectives.
The proposed approach includes using more renewable electricity for AI computing, reducing and recycling water used to cool data centres, and managing computing hardware across its full life cycle.
The ability of these practices to spread will depend partly on how other governments choose to use them.
Zhu Xufeng, dean of Tsinghua University’s School of Public Policy and Management, described two possible routes.
Under the first, governments developing their own AI rules could study China’s policies and adopt the elements that fit their domestic circumstances. Under the second, countries and international organisations could jointly develop frameworks for problems that no government can address alone.
Neither route would involve copying China’s system wholesale.
For many developing countries, Higgins said, the immediate priorities may be basic digital infrastructure, followed by the ability to regulate AI and train people to use it. Countries without the resources to develop an entire regulatory framework could adapt existing solutions from elsewhere.
China’s announcements in Shanghai broadly reflected those priorities.
It pledged 5,000 AI training places for developing countries over five years and proposed international AI application centres with regional groupings covering Southeast Asia, Africa, Arab states, Latin America and Eurasia.
China also said it would deploy Mazu, an AI-powered meteorological warning system, in 30 countries.
Developing economies should not, however, be treated only as recipients of technology and regulatory advice, said Jason Slater, chief AI, digital and innovation officer at the United Nations Industrial Development Organization.
The Global South holds “a huge amount of untapped talent and data,” Slater said. Its unresolved problems could provide valuable AI use cases and regulatory testing grounds.
“We should not underestimate the desire of such countries to make use of AI and to help craft a future dialogue around governance,” he said.
China already plays a major role in AI governance through its research institutions and its ability to convene governments and industry, Higgins said.
The harder test will be implementation.
Certificates must be recognised across borders. Watermarks must remain readable when content moves between platforms. Vehicle and model-testing reports must be trusted by regulators that did not oversee the original tests.
The United Nations’ 2024 Governing AI for Humanity report identified representation, coordination and implementation as persistent weaknesses in the international system. China’s proposals address parts of those gaps, but their credibility will depend on whether they work outside Chinese institutions and markets.
Mongolia may offer an early test.
Former Mongolian president Nambaryn Enkhbayar told the Shanghai forum that his country planned to send students abroad for postgraduate AI training and wanted to cooperate with Tsinghua University and Chinese companies on a digital reconstruction of Karakorum, the ancient Mongol capital.
Slater responded by inviting Mongolia to participate in the first pilot projects under UNIDO’s new industrial AI governance framework.
That exchange illustrates how China’s AI governance experience may initially travel: not as a rulebook exported in full, but as a mixture of training, technical projects and standards tested through cooperation with individual countries.
Whether those early projects develop into broadly accepted international tools remains an open question.
