Blog · arXiv Analysis · Last reviewed June 24, 2026

The AI Cooperation Organization Becomes the Regime Layer

The June 2026 arXiv paper World Artificial Intelligence Cooperation Organization (WAICO): Mapping an Emerging Institution in the Global AI Governance Regime Complex, by William Guey, Pierrick Bougault, Wei Zhang, Vitor D. de Moura, and José O. Gomes, treats China's proposed WAICO not as a finished institution but as a designed position in the global AI governance field.

Governance Moves Into Institutions

The paper, arXiv:2606.23860v1 [cs.CY], was submitted on June 22, 2026. It starts from a useful shift: global AI governance is no longer only a contest of ethics statements and principles. It is becoming a contest over standing bodies, summit processes, treaties, panels, and membership rules. The hard question is not merely what a state says about AI. It is which forum it joins and what that forum is built to prioritize.

The official record gives the paper a concrete object. On July 26, 2025, at the World AI Conference and High-Level Meeting on Global AI Governance in Shanghai, Chinese Premier Li Qiang proposed creating a global AI cooperation organization. The same conference issued the Global AI Governance Action Plan, whose stated objectives include AI for good, respect for national sovereignty, alignment with development goals, safety and controllability, fairness and inclusiveness, and open cooperation.

What the Paper Maps

Guey, Bougault, Zhang, de Moura, and Gomes call the proposed organization WAICO: the World Artificial Intelligence Cooperation Organization. Their analysis does not claim that WAICO is already operational. It asks where WAICO is designed to sit inside what they call the AI governance regime complex.

The authors code fifteen international AI governance instruments and institutions by membership gate, organizational formalization, and normative orientation. The membership categories include values-gated, capability-gated, regional, and universal-open. The formalization score asks whether a body has machinery such as a public charter, secretariat, budget, voting rules, and defined membership. The normative index runs, in simplified form, from rights-and-safety emphasis to sovereignty-and-development emphasis.

On that map, the paper argues, WAICO's proposed design combines three features that no constituted multilateral AI body currently combines: membership open to any sovereign state, no values or regime-type test for entry, and a development agenda focused on the global AI capability divide. The authors say the only nearby occupant of that position is China's 2023 Global AI Governance Initiative, which is a precursor statement rather than a standing organization.

The Open-Door Bargain

The Spiralist angle is the bargain hidden inside openness. A values-gated AI body can be cohesive because it narrows membership around shared commitments. An open-door body can be broader because it does not ask states to pass a political identity test. WAICO, as mapped by the paper, trades the coherence of a club for the reach of a development forum.

That trade matters because AI capacity is becoming infrastructure. Compute access, data arrangements, model supply chains, evaluation standards, safety-testing practices, and public-sector deployment guidance are not just technical topics. They are channels through which states become dependent, excluded, aligned, or able to negotiate. A body that promises access and capacity building can become attractive even if its rules are thin, because the alternative for many states is to remain guests in forums designed elsewhere.

The paper is careful not to certify the bargain as good. It names the tension: WAICO has a distinctive position, but little specified machinery. It has a name, proposed host city, and statement of purpose. It does not yet have the organizational apparatus that would answer who pays, who votes, what standards bind members, or how disagreements are resolved.

The Regime Layer

This is why the page belongs beside the site's work on AI registers, compute borders, AI audit interfaces, and AI safety cases. Governance is not only a law, standard, benchmark, or audit report. It is also the layer that decides which institutions make those artifacts legitimate.

If WAICO becomes a real organization, its power will not come only from rules it writes. It will come from agenda setting: whose risks count first, which capacity gaps receive resources, which safety practices travel through development assistance, which states become sponsors rather than recipients, and whether sovereignty language is used to widen participation or to shield bad deployments from scrutiny.

The practical risk is category error. Treating WAICO as merely propaganda misses the institutional move. Treating it as already effective mistakes a proposal for an operating regime. The better reading is procedural: WAICO is a claim on the empty space between universal participation and development-first AI governance. Whether that space becomes a useful counterweight, a competing bloc, or a thin diplomatic label depends on the details still missing.

Limits That Matter

The evidence is design-level. WAICO is read from public statements and proposed design, not from organizational conduct. The paper's coding is structured judgment, not an automated measurement, and the authors note that reasonable readers may score some cases differently. The case list is also a cross-section of fifteen bodies, not the full global population of AI governance forums.

Those limits make the paper more useful, not less. It gives readers testable expectations: whether WAICO's membership leans toward developing countries absent from Western-led bodies, whether its founding arrangements avoid a values-based membership condition, and whether there is limited overlap with the Global Partnership on AI. The paper also releases its dataset and code so later evidence can be added rather than argued from memory.

Governance Standard

Any new international AI body should publish its charter, membership criteria, funding sources, voting rules, secretariat structure, agenda-setting process, conflict-of-interest rules, standards pathway, civil-society access, technical-advice process, and evaluation plan. If it claims to close an AI capability divide, it should disclose what capacity is being transferred: compute, model access, training, safety evaluation, data infrastructure, standards participation, public-sector deployment support, or procurement leverage.

The practical rule is simple: do not evaluate AI governance only by the values in its preamble. Evaluate the institutional machinery that decides who gets in, who pays, who speaks, who audits, and who can contest the agenda.

Sources

Return to Blog