What was actually signed
What was actually signed is a founding agreement, not a rulebook. On Thursday 16 July 2026, at the World AI Conference in Shanghai, 29 countries signed a document to establish the World Artificial Intelligence Cooperation Organization, or WAICO, an independent intergovernmental body that will be headquartered in Shanghai. Foreign Minister Wang Yi signed on behalf of the Chinese government, and UN Secretary-General Antonio Guterres attended the ceremony.
The named founders lean toward the developing world, with Kazakhstan, Laos, Pakistan, Russia and Indonesia among them, and no EU member state appears on the founding list in any account. Xi Jinping gave the opening keynote, his first in-person WAIC appearance since 2018, pitching China as an AI partner to developing countries, pledging 5,000 AI training places and an "AI-plus" international cooperation initiative built around themes of a "people-centered", "secure and controllable" AI under human oversight.
The real risk is standards fragmentation
Why it matters: the story is not a single ceremony but a second rule track. For years the working assumption has been that AI governance would converge on one broadly Western template, with the EU AI Act as its most detailed expression. A Shanghai-anchored organization with 29 mostly Global-South members gives those states a forum of their own to converge standards around, and it is a forum Europe sits outside.
The timing sharpens the point. The EU AI Act's obligations for general-purpose AI become enforceable on 2 August 2026, only weeks after this signing. Europe is tightening a detailed, binding regime at the exact moment a competing body forms to coordinate a different approach for a large bloc of markets.
For a company deploying AI across those markets, the exposure is fragmentation. Two divergent rule tracks, one anchored in Brussels and one in Shanghai, mean the compliance question is no longer "does this meet the AI rules" but "whose rules, in which market". That is a cost and a design constraint, not a talking point.
The framing is China's, and the caveats matter
Read the caveats plainly. The favorable language around WAICO, that it will be "beneficial to all humanity" and deliver "secure and controllable" AI, is the framing of Chinese state sources such as Xinhua and CGTN, and it should be read as their framing rather than an agreed neutral description. What was signed is a declaration of intent to build an institution, not a body of binding rules that constrains anyone today.
The independent record is narrower and worth holding to. Twenty-nine states signed, the organization will be based in Shanghai, Wang Yi signed for China, Guterres attended, and no EU member joined. Everything beyond that, the mission, the values and the promise of benefit, is aspiration at this stage, and an owner should price it as aspiration.
What owners should do now
The bottom line: in the same week it shipped a frontier open model in Kimi K3, China moved to own the governance architecture of AI, not just the technology. Owning the rules forum is a longer lever than owning any single model, and 29 states now have a place to converge on rules that Europe cannot influence from inside.
The practical response is to stop treating "the AI rulebook" as a single global thing. Map which of your markets are likely to track Brussels and which may drift toward a Shanghai-coordinated approach, then factor that split into vendor selection, model sourcing and where you deploy. A vendor that is clean under one regime is not automatically clean under the other.
None of this requires alarm, and it does not require picking a side. It requires planning for divergence as the base case and building the flexibility to comply with more than one standard. The companies that treat governance fragmentation as a design input now will carry less risk than those that discover it market by market later.
Read next: Germany Now Has a Named AI Regulator | More Than 15 Partners, and No Published Rules
