A parallel institution, not a universal one
On July 16, 2026, representatives from 29 countries gathered in Shanghai to sign the founding agreement of the World AI Cooperation Organization (WAICO), a new intergovernmental body that China's Foreign Ministry says will "promote international cooperation and global governance on AI" (gov.cn). Foreign Minister Wang Yi signed for Beijing; UN Secretary-General António Guterres attended. The founding members — Russia, Belarus, Serbia, Cuba, Brazil, Venezuela, Kazakhstan, Pakistan, Indonesia and roughly 20 more African and Asian states — have one thing conspicuously in common: none belong to the G7, and none are EU member states (Al Jazeera).
The next day, President Xi Jinping gave the WAIC 2026 opening keynote in person for the first time, using the moment to unveil concrete commitments: 5,000 AI training slots and seminars for developing-country officials over five years, new cooperation centers with ASEAN, the African Union and the Arab League, and deployment of China's MAZU AI weather-warning system to 30 countries (gov.cn). Xi framed it as building "a symphony of international cooperation" rather than a solo performance by any single power.
The steelman: the Global South has a real grievance
Before dismissing WAICO as pure geopolitics, the underlying complaint deserves a fair hearing. The AI governance conversations that have mattered most over the past three years — the UK's Bletchley Park summit, the Hiroshima AI Process, the EU AI Act, the US executive orders on AI safety — were negotiated almost entirely among wealthy, English- or European-language-speaking capitals. Compute, foundation models and frontier-lab talent are concentrated in the US and China; the countries most likely to be governed by AI systems built elsewhere have had the thinnest seat at the table. Caixin's reporting notes that WAICO explicitly operates outside the UN system, positioned to move faster than UN processes that Guterres himself has called for supplementing with "a specific global development fund" for AI (Caixin). A body that puts 5,000 training slots and a shared weather-forecasting tool on the table for Indonesia, Ethiopia or Pakistan is offering something tangible that Western summits, heavy on communiqués and light on capacity transfer, largely have not.
The catch: governance without governors
But the steelman only carries WAICO so far, because the structure undercuts the stated mission. WAICO is a strictly intergovernmental body — states signing agreements with states, headquartered in a city whose AI sector answers to the Cyberspace Administration of China's algorithm-registration and content-alignment regime. There is no announced mechanism for independent researchers, civil society groups or the companies actually building frontier models to participate, unlike multistakeholder efforts such as the Global Partnership on AI. That matters because the governance model a country imports along with training slots tends to stick. A member state that adopts Chinese-built AI curricula, Chinese-standard weather-AI infrastructure and a seat in a China-chaired secretariat is also, functionally, importing assumptions about state control over algorithms and information flows — the same assumptions that gave the CAC unusually broad discretion to fast-track or freeze deployment of generative AI products through China's own registration filings, most recently expanded in the 2023 Interim Measures for Generative AI Services. Xi's own framing — calling for governments to avoid "overstretching the national security concept" while simultaneously building an institution to "establish regulations and monitoring systems" — is a pointed jab at US export controls, but it doesn't answer who checks WAICO's own monitoring powers, since there is no independent oversight body written into the founding agreement as reported.
Why fragmentation is the actual risk
The deeper problem for an open internet isn't that China built a club — it's that global AI governance is now splitting into two non-interoperable blocs rather than converging on shared technical baselines. A startup training a model for deployment across both WAICO states and the EU AI Act's jurisdiction will increasingly face two incompatible compliance regimes, negotiated in two capitals that don't talk to each other. That raises costs for exactly the kind of cross-border AI diffusion that has driven the most benefit to lower-income countries — the opposite of WAICO's stated goal of narrowing the access gap. Duplication also drains attention from the UN's own Independent International Scientific Panel on AI and Global Dialogue on AI Governance, established by General Assembly resolution in 2024, which at least in principle answer to all 193 member states rather than a self-selected founding cohort.
The proportionate response
None of this justifies Western governments treating WAICO as a threat to be blocked rather than a signal to act on. The honest reading is that the US, EU and allied donors ceded this diplomatic space by treating AI capacity-building as an afterthought to their own safety-summit theater. The proportionate response is not to match Beijing bloc-for-bloc but to make the case, with comparable resources, for governance built on multistakeholder input and independent oversight rather than government-to-government discretion — and to keep pushing, through the UN's own AI bodies, for the interoperable technical standards that a two-bloc world is now put at risk.