Twenty-nine countries founded an AI organisation in Shanghai. The agreement does not say what it can do.

On 16 July, in Shanghai, twenty-nine countries signed an agreement establishing the World Artificial Intelligence Cooperation Organization. Foreign Minister Wang Yi signed for China. The Secretary-General of the United Nations, António Guterres, was in the room. The following morning Xi Jinping opened the World AI Conference with a keynote titled "Joining Hands to Build a Just and Equitable System for Global AI Governance," and China's foreign ministry described the new body as "a major move by China to answer the call of the Global South" and "an important milestone in the history of AI development."

The reported version, which has now circled the world several times, is that China has launched a rival bloc to govern artificial intelligence. That version is not wrong, exactly. It is simply ahead of the paperwork.

Because the useful question about any new institution is never whether it has been founded. It has been; the signatures are real. The question is what its constituting text requires, of whom, and by when. On that question the documents released so far are close to silent, and the silence is the story.

What the agreement says

The operative language, as carried by Xinhua and the State Council, is short enough to quote in full effect. WAICO is to be "an independent intergovernmental international organization headquartered in Shanghai." It "aims to promote international cooperation and global governance on AI, ensuring that AI is beneficial, safe and fair." It "will uphold the purposes of the UN Charter, be committed to extensive consultation and joint contribution for shared benefit and adhere to a people-centered approach."

The Chair's Statement issued at the close of the high-level meeting goes further in ambition, calling WAICO "the world's first intergovernmental international organization on AI" and adopting fifteen principles for global AI governance under the heading "AI Partnership for a Brighter Future." It also, notably, endorses "the central role of the U.N. in global governance" and expresses support for the UN's own Global Dialogue on AI Governance.

Now the absence. Across the Xinhua release, the State Council's account, the foreign ministry readout, the Global Times report and the Chair's Statement, there is no description of a governing council or general assembly. No secretariat. No voting rules or decision procedure. No budget, no schedule of contributions, no financing model. No accession criteria and no withdrawal clause. No compliance, dispute-resolution or enforcement mechanism of any kind.

This needs stating carefully, because the distinction matters. None of those sources says WAICO lacks these things. They simply do not describe them. It is entirely possible that a fuller charter exists and has not been published in English, or that the apparatus is to be negotiated among members over the coming year. That is ordinary. Institutions are often founded before they are constituted.

But until that text appears, the honest description of WAICO is the one a group of researchers reached in a preprint written before the signing, when they scored it against fifteen comparable institutions and placed it last on formalisation: it has a name, a host city and a statement of purpose, and the apparatus that would let it act has not yet been specified. That assessment pre-dates the founding and has not been peer-reviewed, and should be read with both caveats attached. It has also not yet been contradicted by anything China has published. The Simon Institute for Longterm Governance, surveying the same material, reports simply that there is no public information on a budget or financing model.

Institutions are often founded before they are constituted. The question is how long the gap lasts, and who benefits while it does.

Who signed, and who did not

Chinese state accounts name only a handful of signatories directly — Kazakhstan, Laos, Pakistan, Russia and Indonesia among them. The Global Times describes the composition as including Russia, Belarus, Serbia, Cuba, Brazil and Venezuela, plus ten African and twelve Asian countries. A full roster of the twenty-nine has circulated in secondary sources; I have not been able to confirm it against a primary Chinese-government publication, and it should be treated accordingly.

What can be said from the membership as published is narrower and still significant. The United States is not a founding member. Neither is any EU member state, nor the United Kingdom, Japan, India, South Korea, Canada or Australia. This is a founding of the Global South and the states already outside the Western technology-governance orbit, and China has been explicit that this is the intended constituency rather than an accident of who showed up.

Xi's speech was correspondingly pointed without naming anyone. "AI development should not be a solo performance by a single country, but a symphony of international cooperation," he said, and "we must jointly oppose overstretching the national security concept in the field of AI and placing one country's security over that of others." He committed China to providing developing countries with 5,000 AI training and seminar opportunities over five years, to building application cooperation centres with ASEAN, the Arab League, the African Union, CELAC, the SCO and BRICS, and to enabling thirty countries to use its AI-powered meteorological warning system.

Those are capacity-building commitments, not regulatory ones, and they are the most concrete things in the entire package. That tells you what kind of body this is designed to be, at least at the outset. It offers training, tooling and access. It does not, on the present record, ask anyone to do anything.

The lineage, and the twelve-month gap

None of this arrived suddenly. Xi proposed a Global AI Governance Initiative in October 2023. At last year's World AI Conference, China proposed a global AI cooperation body and floated Shanghai as its seat. Xi repeated the proposal at APEC in October 2025. The agreement was signed twelve months after the proposal.

Twelve months from proposal to signature is fast for an intergovernmental organisation, and speed of that kind usually has a price. The price is normally paid in the constituting text — the parts that require members to agree on money, votes and obligations are the parts that take years, and they are precisely the parts absent here. A treaty that binds no one is quick to draft. That is not a criticism of the drafters; it is an observation about what has been achieved so far, which is convening rather than commitment.

What a binding regime looks like, on the same calendar

It is worth setting WAICO beside the instrument it is most often contrasted with, because the comparison is instructive in both directions.

On 2 August 2026 — two weeks from now — the majority of the EU AI Act's rules come into force and enforcement begins. The Commission's own implementation timeline is specific: the transparency obligations in Article 50 start to apply, measures supporting innovation start to apply, and "enforcement of the AI Act starts at national and EU-level concerning general-purpose AI models, prohibitions, transparency rules and AI literacy." There are designated national competent authorities. There is an AI Office. There are conformity assessments, a registration database and administrative fines calculated against global turnover.

That is what binding looks like: a date, a named enforcer, a defined class of addressee, and a consequence.

And here is the part European officials are less keen to foreground. The Act's hardest obligations — the Annex III rules for high-risk AI systems — now apply from 2 December 2027, deferred by roughly sixteen months under this year's simplification package. A further transitional deadline of 2 December 2026 covers new prohibitions on deepfakes and child sexual abuse material, and certain synthetic-content provisions.

So the honest ledger for the fortnight ahead reads: the world's most binding AI statute is slipping its most demanding deadline by sixteen months, and the world's newest AI organisation has no deadline to slip. Neither of those facts flatters anyone. They are different failure modes of the same underlying difficulty, which is that nobody has yet worked out how to write enforceable rules for a technology that changes faster than a legislative cycle.

The third model, which is also not a treaty

The United States governs AI internationally through export controls and a coalition rather than through an organisation. Its vehicle is Pax Silica, the State Department's flagship effort on AI and supply-chain security, launched in December 2025 and now carrying twenty-four signatories including Japan, South Korea, India, Israel, the UK, the UAE and, since this year's summit, the European Union and Germany.

Pax Silica is not a treaty either. It conditions access to chips and tooling, which is a real instrument with real effects, but it binds through commercial dependency rather than through law. That is worth naming plainly, because the comparison usually offered — binding West, unbinding East — does not survive contact with the documents. There are currently three international approaches to AI governance and only one of them, the European, actually issues obligations to named parties on published dates.

One detail captures how little these blocs are actually blocs. Kazakhstan is a founding member of WAICO and a signatory to Pax Silica. States are not choosing sides. They are joining everything that offers something and committing to nothing that costs anything, which is rational behaviour when none of the instruments yet imposes a duty.

What to watch, and what convening power is worth

It would be a mistake to conclude that a body without enforcement powers is therefore inconsequential. The Brussels Effect works through market access — Europe conditions entry to a large wealthy market, and firms comply globally because building two products is expensive. WAICO has no comparable market to condition. What it has instead is a seat at the table where technical standards, capacity-building programmes and model norms get drafted, and a constituency of states that have not yet written domestic AI law and will need a template when they do. Standards adopted early by thirty countries that are still drafting tend to become the default for those countries permanently. That lever is slower than market access and, over a decade, not obviously weaker.

There is also the question of the United Nations, where the position is genuinely ambiguous. The UN held the first session of its own Global Dialogue on AI Governance in Geneva on 6 and 7 July, nine days before the Shanghai signing, and its Independent International Scientific Panel on AI was seated in February. Guterres attended the WAICO ceremony and spoke at the conference the next day, telling the room that "technology that will shape the future of humanity must be shaped by all of humanity" and that "every nation needs a seat at the table." The UN's own account of his Shanghai remarks does not mention WAICO once. Whether the new organisation ends up complementing the UN process or competing with it for the attention of the same developing states is not answerable from the founding documents, because the founding documents do not address it.

Analysts have read the timing as opportunistic. Arindrajit Basu of the Carnegie Endowment put it directly: "With Washington rapidly retreating from global cyber and AI norms-setting processes and withdrawing its financial backing for cyber diplomacy more broadly, Beijing is keen to demonstrate its global leadership." As of Sunday, Washington had made no public response to the founding at all.

The load-bearing date, then, is not 16 July. It is whenever WAICO publishes a charter with a decision rule in it. Until an organisation can say who votes, who pays and what happens to a member that ignores it, what has been created is a forum with a building. Forums matter — they set agendas, and agendas eventually become texts. But a forum is not a regime, and the difference will be visible in the first document that tells a member state it must do something.

Read that document when it comes. Not the communiqué announcing it.