China’s new AI-governance bloc turns standards into a geopolitical contest

Reporting

Twenty-nine countries signed an agreement in Shanghai on July 16 to establish the World Artificial Intelligence Cooperation Organization, a new intergovernmental body headquartered in Shanghai and presented by China as a venue for global cooperation on artificial intelligence. That is the confirmed development: not a model launch, not another voluntary company pledge, but a new diplomatic institution meant to compete for authority over how AI rules, standards, capacity-building, and safety norms are written.

The launch matters because AI governance is moving from conference language into institutional architecture. The European Union has the AI Act. The Group of Seven has pushed voluntary model-risk commitments. The United Nations has tried to keep the conversation universal through advisory work. The United States has leaned toward a more innovation-centered approach, with national-security controls and procurement rules doing much of the hard governing work. China is now offering another pole: a Shanghai-based body aimed especially at countries that do not want the rules of the AI economy written only in Brussels, Washington, or frontier-model boardrooms.

Chinese state media and the Chinese government’s English portal report that the founding members include Kazakhstan, Laos, Pakistan, Russia and Indonesia. Reuters reported a broader roster including Russia, Belarus, Serbia, Cuba, Brazil and Venezuela, plus 10 African and 12 Asian countries. United Nations Secretary-General António Guterres attended the signing ceremony, according to those accounts, though attendance should not be confused with UN ownership of the organization or endorsement of every policy position that may emerge from it.

The organization’s stated purpose is broad and appealing: beneficial, safe and fair AI; global cooperation; a people-centered approach; and adherence to the purposes of the UN Charter. No serious reader should dismiss those goals just because Beijing is promoting them. Countries with less domestic compute, less cloud bargaining power, fewer frontier labs and weaker leverage over dominant platforms have real reasons to worry that AI rules could be written around the interests of rich states and already-powerful companies.

But no serious reader should treat the launch as neutral technocracy either. A global AI body headquartered in Shanghai, proposed by China and launched with China’s foreign minister signing on behalf of Beijing, is also an instrument of influence. It arrives as governments are deciding whose safety tests count, whose model evaluations travel across borders, which open-weight systems are treated as public goods, which export controls are legitimate, and whether “AI for development” becomes a capacity-building project or a dependency channel.

The immediate significance is not that WAICO now controls global AI policy. It does not. The public record available so far does not establish binding enforcement powers, funding commitments, voting rules, inspection authority or a compliance mechanism. The full founding agreement has not yet been widely published in English. Those are material unknowns.

What each side is trying to prove

China’s pitch is easy to understand. It wants to be seen not merely as a manufacturing and model-building power, but as a rulemaking power. At the World Artificial Intelligence Conference in Shanghai, Chinese President Xi Jinping used the moment to frame China’s approach around open-source or open-weight AI, access for developing countries, safety controls, and warnings against new inequalities in the AI era, according to Reuters. That frame lets Beijing contrast itself with a U.S. ecosystem dominated by closed frontier labs, high-cost cloud infrastructure and export-control policy.

For countries outside the richest technology blocs, there is a practical appeal. If a government wants local-language models, public-sector AI training, cheaper deployment, or leverage against U.S. cloud vendors, Chinese labs and Chinese-backed standards forums may look less like ideology and more like procurement optionality. The Global South is not a prop in this story. Many governments are trying to build administrative capacity, education tools, health systems, disaster-response models and industrial policy without being permanently downstream of U.S. platforms or EU compliance templates.

The counterargument is also strong. Open access is not the same as democratic governance. A standards body can distribute tools while normalizing surveillance-heavy state use, weak civil-liberties protections, or vague safety language that sounds universal but avoids enforceable rights. China’s domestic technology-governance model includes real state direction over platforms, data and speech. Countries concerned about censorship, civil society oversight, independent audits and judicial remedies will want to know whether WAICO will make room for those constraints or treat them as Western preferences.

The U.S. and Europe have credibility problems of their own. When Washington speaks about AI safety, it often does so through national-security controls, procurement requirements and relationships with a few dominant private labs. When Europe speaks about rights-based AI governance, it also exports compliance costs that smaller firms and less-resourced governments may struggle to absorb. The EU AI Act is the most developed statutory framework among major economies, but it is not a development-finance plan, a compute-access program, or a guarantee that non-European governments will have equal influence over standards.

The policy consequences

The first consequence is fragmentation. Companies building AI systems for global use may face not one international conversation but several overlapping ones: EU legal obligations, U.S. export-control and security expectations, voluntary frontier-model commitments, UN recommendations, sector-specific rules, and now a China-led intergovernmental organization claiming space in global governance. That does not automatically mean chaos. Overlapping institutions are normal in technology policy. But “global standard” will increasingly be a contested label, not a settled fact.

The second consequence is pressure on the United Nations. The UN’s High-level Advisory Body on AI has argued for inclusive governance and warned that AI’s benefits and risks are unevenly distributed. WAICO borrows some of that vocabulary while moving convening power to Shanghai. If the UN wants to remain the universal venue, it will need more than careful reports. It will need mechanisms that developing countries see as materially useful: capacity, technical assistance, evaluation access, compute partnerships and a voice in standards that is not purely ceremonial.

The third consequence is pressure on Washington and Brussels to distinguish principled governance from market protection. Some Chinese claims about open access will be self-serving. Some Western claims about safety will be self-serving too. Export controls may be justified by military or national-security concerns; they may also be perceived abroad as an effort to preserve U.S. commercial dominance. A durable AI order needs to answer both the safety question and the access question.

What remains unverified

Several important facts remain unresolved. The full founding agreement, if published, should be read before anyone claims WAICO has binding authority. The membership list should be checked against official statements from each founding government, not only host-government or wire-service summaries. Funding, staffing, voting structure, dispute procedures, technical working groups, industry participation, civil society access and audit standards are not yet clear from the public accounts reviewed for this column.

It is also too early to say whether WAICO will produce meaningful safety work. The language of “safe and fair” AI is now common across nearly every faction in the debate. The test is whether a body can specify risk thresholds, publish evaluation methods, protect affected people, handle incidents, and resist capture by either governments or vendors. A standards forum that cannot say no to its most powerful sponsor is not a regulator. It is a stage.

Analysis

The disciplined way to read WAICO is neither panic nor applause. It is a bid for governance power at a moment when AI power is being concentrated in chips, cloud contracts, model weights, standards bodies and public-sector procurement. China is telling the world that AI rules should not be written solely by the United States, Europe and a handful of frontier labs. That argument will land in many capitals because it contains a truth. It will also raise hard questions because the messenger has its own record, interests and red lines.

For readers trying to track what changes next, ignore the ceremonial language and watch the plumbing: who signs the final instrument, who funds the secretariat, whether technical standards are published, whether independent researchers and civil society can participate, whether safety testing is reproducible, whether member governments start referencing WAICO in procurement or domestic AI laws, and whether companies use WAICO language to market models abroad.

If those things happen, today’s signing will look less like summit choreography and more like the start of a parallel AI-governance system. If they do not, it will be another grandly named institution in a field already crowded with them. Either way, the policy lesson is plain: AI governance is becoming multipolar, and the fight is not just over rules. It is over legitimacy.

Sources

• Chinese government English portal: “29 countries sign agreement on establishing World AI Cooperation Organization”
• Xinhua: “Update: 29 countries sign agreement on establishing World AI Cooperation Organization”
• Reuters: “Twenty-nine countries sign agreement to establish global AI cooperation body”
• Reuters: “Xi pitches China as leader of new global AI order, challenging US dominance”
• European Commission: AI Act overview and application timeline
• United Nations: High-level Advisory Body on Artificial Intelligence

---

Shadowfetch is a technology publication. Explore Shadowfetch Linux — our own Linux build — and the Shadowfetch apps on the App Store.