Shanghai AI pact gathers 27 states, putting Beijing at the centre of a parallel governance track

Russia, China and more than 25 other countries put their signatures to a single page in Shanghai on Thursday, 16 July 2026, committing to establish the World Artificial Intelligence Cooperation Organization. The signing ceremony, hosted in the Chinese financial hub, sets up the institutional skeleton of a Beijing-centred AI governance track running alongside the Western-led processes that have, until now, set most of the rules of the road for frontier models.

The new body's political weight will depend on three variables the ceremony did not resolve: who funds it, whose technical standards it codifies, and whether Western capitals treat it as a competitor or ignore it. The text circulated on Thursday is aspirational rather than operational, but the membership list, with Moscow and Beijing at its core and more than two dozen states flanking them, signals that the era of a single global conversation about AI governance is closing.

What was actually signed

CGTN's official account on X described the event as "a signing ceremony of the Agreement on the Establishment of the World Artificial Intelligence Cooperation Organization," held in Shanghai on Thursday. Tasnim News Agency, an Iranian outlet with close ties to Tehran's foreign policy establishment, reported the same fact in parallel: "Russia, China and more than 25 other countries signed an agreement to establish the Artificial Intelligence Cooperation Organization."

Neither release published the full text of the agreement, the legal personality of the new body, its secretariat structure, or its funding mechanism. The phrasing in both wire summaries refers to an "agreement on establishment," the standard diplomatic register for treaties that create an international organisation, leaving the binding force of the document slightly ambiguous. The signatories appear, on the public record, to have committed to a process rather than to a finished institution.

That process is the news. Twenty-seven or more states, by Tasnim's count, agreed in a single room to coordinate on AI under a framework whose centre of gravity is unmistakably in Shanghai rather than in Brussels, Washington, or London.

The counter-narrative from the West

Western wire reporting on AI governance over the past eighteen months has centred on the AI Safety Summit process, the EU AI Act, the US AI executive-order cycle, and the Bletchley-to-Seoul-to-Paris summit track. That architecture has produced concrete instruments: a binding risk-classification regime in Brussels, voluntary compute thresholds announced in Washington, frontier-model evaluation commitments from a dozen Western labs.

The Shanghai pact reads as a deliberate alternative to that track rather than an addition to it. By staging the ceremony in Shanghai and using language that foregrounds "cooperation" rather than "safety" or "risk," the organisers have offered Global South capitals a seat at a table of their own rather than observer status at a Western one. The framing choice is not incidental. "Cooperation" implies shared development; "safety" implies restriction. For governments facing domestic pressure to industrialise rather than restrain, the Shanghai register is the easier one to sign.

The Western counter-line, where it has begun to form, treats the pact as a soft-power vehicle for Chinese technical standards and a diplomatic shield for Russian access to compute and chips that would otherwise be constrained by sanctions. That reading has internal logic. But it also flattens the agency of the two dozen-plus non-Russian, non-Chinese signatories, several of which have active AI industrial policy of their own and would object to being cast as junior partners in someone else's project.

What the membership actually tells us

Tasnim's reporting frames the pact as a Russian-Chinese bilateral with a large supporting cast. The headline construction, "Russia, China and more than 25 other countries," gives Moscow co-equal billing with Beijing, which is consistent with how both governments have chosen to project the event. The CGTN framing, by contrast, centres Shanghai as the host and the agreement as a multilateral product, with Moscow recast as a prominent signatory rather than a co-author.

That asymmetry is itself a clue. Beijing controls the venue, the secretariat logistics, and the technical-staffing pipeline that any AI cooperation organisation will require. Moscow contributes diplomatic weight and a sanctions-circumventing rationale for the body, since a Chinese-Russian joint platform is structurally harder for Western export-control regimes to reach than either country acting alone. The other twenty-five-plus signatories are the political floor: enough mass to claim multilateral legitimacy, drawn from regions where the Western AI governance track has the weakest traction.

The sources do not specify which countries outside Russia and China signed, what reservations were recorded, or whether any major AI-capable state outside the China-Russia orbit, India, Indonesia, Brazil, Saudi Arabia, the United Arab Emirates, has formally joined. That omission is the most consequential gap in the public record at this stage.

Stakes over the next eighteen months

If the new organisation survives its first year and produces even a non-binding technical standard, a model-evaluation protocol, or a compute-disclosure template under its own imprimatur, the practical effect will be the emergence of two parallel compliance regimes for any AI lab that wants to sell into both blocs. Labs will face a choice: certify to the Shanghai process, the Brussels process, or both, and pay the engineering overhead of operating inside two regulatory worlds.

For Beijing, the upside is legitimacy as a standard-setter rather than a standard-taker. For Moscow, the upside is friction in the chip and compute export-control architecture built since 2022, since a multilateral body provides cover for technology transfers that bilateral arrangements cannot. For Global South signatories, the upside is leverage: a seat in a forum where their domestic industrial-policy priorities are not pre-filtered through Western risk taxonomies.

The losers are the Western forums themselves, in proportion to how much real work they were already doing. The AI Safety Summit track produced communiqués; the EU AI Act produced law. The Shanghai pact has produced, on Thursday, a piece of paper and a photograph. The contest over the next eighteen months is whether the piece of paper fills out into an institution capable of issuing its own technical standards, or whether it remains a diplomatic trophy that complicates Western coordination without displacing it.

What the sources do not specify is the budget, the headquarters lease, the staff complement, or the timetable for the body's first substantive output. Until those numbers exist, the Shanghai pact is a flag planted rather than a building opened.

This article tracks an emerging multilateral track rather than a finished institution; the article's sourcing will be expanded as the agreement text and the full signatory list are made public.

Wire provenance

This editorial synthesis draws on the following public wire/social posts:

• https://t.me/s/JahanTasnim