The Rulebook and the Referee: AI Governance Splits in Two
China's WAICO writes the rules while America's Pax Silica controls the pipes -- and neither architecture has a seat for the governed. A democratic-legitimacy audit of the week AI governance split in two.
On 16 July 2026, in Shanghai, representatives of 29 countries signed the founding agreement of the World Artificial Intelligence Cooperation Organization -- WAICO -- the newest intergovernmental organisation on Earth, and described by its founders as the first devoted to artificial intelligence [1][16]. China's foreign minister, Wang Yi, signed for the host. The United Nations Secretary-General, António Guterres, attended the ceremony. The next day, Xi Jinping appeared at the World AI Conference for the first time in the event's history and hailed the newly created body as "an important milestone in the history of AI development", urging that AI be "a symphony of international cooperation" rather than "a solo performance by a single country" [2].
Look down the list of the 29 founders -- Russia, Brazil, Indonesia, South Africa, Pakistan, and roughly ten African and twelve Asian states among them -- and one absence organises everything else: not a single G7 or European Union member signed [1][16]. The world's newest international organisation was born without one established industrial democracy in the room.
That fact can be read two ways, and both readings are correct. The first: China has consolidated a bloc. The second: the democracies never made an offer these 29 governments preferred.
Pipes and rules
Seven months before Shanghai, in December 2025, the United States State Department convened its own architecture: Pax Silica, a supply-chain alliance covering semiconductors, AI infrastructure, and rare earths. By July 2026 its signatories included the United Kingdom, Japan, South Korea, Australia, the Netherlands, the European Union, and -- notably -- India, which signed in February [4].
The two structures are not mirror images; they are different organs of power. The analyst George Chen put the distinction plainly: "Pax Silica is about the 'pipes' of AI, while WAICO is about the 'rules of the game.' Washington wants to control the backbone; Beijing wants to control the norms" [5]. One bloc secures who gets the chips. The other proposes who writes the standards and trains the regulators -- and so who defines what "safe" and "fair" mean for the majority of humanity. India's former permanent representative to the United Nations, Syed Akbaruddin, sharpened that second ambition months before the launch: when finance was globalised, he observed, the West wrote the rules at Bretton Woods -- now Beijing seeks "both the rulebook and the referee under its roof" [15].
The launch imagery, however, outran the institutional substance. The organisation has, as of this writing, no published charter text in English, no secretary-general, no disclosed budget, and no organs of decision-making. An academic mapping of fifteen AI governance bodies, published in June 2026 by researchers at Tsinghua University and the Federal University of Rio de Janeiro, ranked WAICO lowest of all fifteen on formalisation -- in the authors' words, "a name, a host city, and a statement of purpose" [6]. Even the headline pledges announced alongside the launch -- 5,000 AI training placements for developing countries over five years, cooperation centres with regional groupings from ASEAN and the African Union to the Arab League and BRICS, and access for 30 countries to China's MAZU meteorological AI system -- are national commitments by China, not programmes of the new organisation [2][7].
A hollow institution, then? The history of institution-building counsels caution. Founding documents are thin by nature; what matters is whether machinery accretes around them. WAICO's declared pillars -- capacity building, governance frameworks, technical standards coordination, industrial cooperation -- are exactly the machinery that, if built, shapes markets without ever passing a binding law [8]. The precedent is China's Digital Silk Road, where infrastructure arrived first and standards followed the installed base. As Arindrajit Basu of the Carnegie Endowment has documented, Chinese firms already hold critical positions in the AI stacks of Southeast Asia, Africa, and Latin America, with offerings frequently cheaper than Western alternatives [9].
Why they signed
The temptation in Western capitals will be to explain the 29 signatures as coercion or clientelism. The evidence points somewhere less comfortable: the offer was better.
Consider what a developing-country government weighing its AI future sees in mid-2026. Chinese open-weight models -- DeepSeek, Alibaba's Qwen, and their derivatives -- accounted for roughly 61 per cent of token volume on OpenRouter, a widely used model-routing platform, by May 2026, according to one widely cited analysis of platform data; Qwen alone has passed one billion cumulative downloads on Hugging Face, and about 40 per cent of new model derivatives on the platform are now built on it [10]. Malaysia, a WAICO founding member, has announced that its sovereign AI ecosystem will run on Chinese open-weight models [10]. This is capability that is free to download, legal to modify, and indifferent to the adopter's human-rights record. WAICO's membership rules match the models: open to all sovereign states, no values test [8].
Now consider the rival offer. The United States spent 2026 tightening export controls -- Nvidia's own annual report describes the company as "effectively foreclosed" from China's data-centre market, and the restrictions' logic is exclusion, extended in June to Chinese firms' overseas subsidiaries [11]. The European Union, keeper of the rights-anchored regulatory model, deferred the AI Act's high-risk obligations in June 2026 by sixteen to twenty-four months, in part because the supporting technical standards were not ready -- and joined Pax Silica the same month [12]. Washington, meanwhile, has been retreating from the normative field: withdrawal from the Freedom Online Coalition, reduced cyber-diplomacy budgets, and an AI Action Plan that frames international engagement chiefly as countering Chinese influence [9].
A government in Nairobi or Jakarta is not choosing between tyranny and liberty when it signs the WAICO agreement. It is choosing between an organisation offering training slots, model access, and help with standards, and a rights-based order whose most visible recent offerings have been export controls, a postponed statute, and summit communiqués. That comparison undersells what the democracies actually have -- rule of law, recourse, institutions that outlast their governments -- but governments procure on what is visibly on the table. Indonesian technology analyst Heru Sutadi has argued that the real danger is neither bloc winning but an "AI divide" that leaves developing countries structurally behind under either dispensation [3]. The 29 signatures suggest many governments have decided that, divide for divide, Beijing's terms are the ones on the table.
The squeezed commons
There is, formally, a third architecture: the United Nations. Its Global Dialogue on AI Governance held its inaugural session in Geneva on 6 and 7 July 2026 -- ten days before the Shanghai signing -- where the new Independent International Scientific Panel on AI, co-chaired by Yoshua Bengio and Maria Ressa, presented its first report, released days earlier [13]. The UN track is the only forum where membership is universal and the poorest state holds the same formal standing as the largest.
Both blocs pay it tribute. China's founding documents describe WAICO as supporting "the UN as the main channel" for AI governance, and Guterres's presence at the signing lent that claim visibility [8]. Yet Carnegie's analysis suggests a more instrumental reading: a standing Global South coalition, convened in Shanghai, is also a voting bloc that can be marshalled inside UN processes [9]. The democracies are not obviously better custodians: the New Delhi Declaration that emerged from India's AI Impact Summit in February 2026 -- endorsed by some ninety countries and organisations, including both the United States and China -- did not so much as reference the UN Global Dialogue. As Konstantinos Komaitis of the Atlantic Council's DFRLab observed of the summit ecosystem, parallel initiatives "risk evolving in isolation," producing declarations without institutions, budgets, or enforcement [14].
The commons, in short, is being praised by all parties and built around by all parties.
Where is the demos?
The Global Federation's interest in this story is not the horse race. It is the constitutional question underneath it: by what right, and with what accountability, will artificial intelligence be governed?
Audit both architectures honestly and the answer is bleak. WAICO is an agreement among executives -- many of them unelected -- with no published charter, no parliamentary ratification debate of record in most signatory states, and no mechanism by which a citizen of a member country could learn what was agreed in her name, let alone contest it. Pax Silica is scarcely more participatory: a supply-chain compact negotiated by trade and foreign ministries, structured around corporate export interests, with no meaningful public consultation in any signatory democracy. The EU's AI Act -- the one instrument on the field that emerged from an elected legislature, however imperfectly -- is the instrument that just slipped its own deadlines.
The pattern should be named: AI governance is being constitutionally malformed at birth. Rules for the most consequential technology of the century are being written in executive session, on both sides of the divide, with publics treated as downstream recipients rather than upstream principals. The 29 governments that signed in Shanghai did not poll their citizens; neither did the roughly two dozen that have signed the American pact.
If there is a hopeful reading, it lies with the states refusing to enlist. Kazakhstan has signed both WAICO and Pax Silica. Brazil and Indonesia are WAICO founders that remain engaged with Western frameworks; India signed the American hardware pact while staying out of both blocs' rule-making bodies and convening its own summit, whose declaration both superpowers endorsed [4][14]. The Tsinghua-Rio de Janeiro researchers describe the emerging landscape not as two blocs but as an overlapping "regime complex" in which states multi-home and hedge [6]. Multi-homing is not democracy. But it is agency -- evidence that most of the world declines to be conscripted, and that the architecture remains plural enough to contest.
What democrats should do
Three conclusions follow for anyone who believes governance should answer to the governed.
First, take WAICO seriously without taking it at face value. Today it is a nameplate. The tripwires that would make it consequential are observable: a published charter, a named secretary-general, a first technical standard, a funded compensation or compute-access mechanism. Democracies should decide their response to each tripwire before it fires, not after.
Second, compete on the offer, not the lecture. The 29 signatures are a market verdict on the West's proposition to the developing world. A credible democratic alternative would couple compute access and open capability with rights -- not condition the former on geopolitical alignment while delivering the latter as commentary.
Third, defend the one universal room. The UN track is slow and easily mocked. It is also the only architecture in which governance could even in principle be accountable to all of the governed. Letting it become the fig leaf both blocs gesture toward while institutionalising elsewhere would convert a thin commons into an empty one.
The technology at stake concentrates power faster than any previous object of international law. The institutions forming around it, on present evidence, concentrate power too -- merely in different capitals. Whether the governed get a seat is not a question either Shanghai or Washington is currently asking. Someone should.
Sources
[1] Xinhua, "29 countries sign agreement to establish World Artificial Intelligence Cooperation Organization," 16 July 2026. https://english.news.cn/20260716/2d364f232dd7456c8edf1d67d5125d37/c.html [2] Ministry of Foreign Affairs of China, Xi Jinping keynote to the 2026 World AI Conference, 17 July 2026. https://www.mfa.gov.cn/eng/xw/zyjh/202607/t20260717_11984910.html [3] Wikipedia, "World Artificial Intelligence Cooperation Organization," accessed 21 July 2026 (founding member list; Sutadi commentary). https://en.wikipedia.org/wiki/World_Artificial_Intelligence_Cooperation_Organization [4] U.S. Department of State, "Pax Silica Initiative," December 2025; Wikipedia, "Pax Silica," accessed 21 July 2026. https://www.state.gov/releases/office-of-the-spokesperson/2025/12/pax-silica-initiative/ ; https://en.wikipedia.org/wiki/Pax_Silica [5] George Chen, "AI diplomacy divides the world," 16 July 2026. https://georgechen.substack.com/p/ai-diplomacy-divides-the-world-us [6] Guey et al., "Mapping the AI governance regime complex," arXiv, June 2026. https://arxiv.org/html/2606.23860v1 [7] Sixth Tone, "Xi pledges AI training for developing countries at WAIC," 17 July 2026. https://www.sixthtone.com/news/1018788 [8] ChinaCase, "What is WAICO? A guide to the new organization," 19 July 2026. https://chinacase.substack.com/p/what-is-waico-a-guide-to-the-new [9] Arindrajit Basu, "China's Pivot on Global AI," Carnegie Endowment for International Peace, 21 May 2026. https://carnegieendowment.org/research/2026/05/chinas-pivot-on-global-ai [10] Data Gravity, "China's open-weight takeover," 2026. https://www.datagravity.dev/p/chinas-open-weight-takeover [11] Nvidia Corporation, Form 10-K for fiscal year 2026, filed January 2026; Al Jazeera, "US says ban on AI chip shipments applies to Chinese firms outside China," 1 June 2026. https://www.sec.gov/Archives/edgar/data/1045810/000104581026000021/nvda-20260125.htm ; https://www.aljazeera.com/economy/2026/6/1/us-says-ban-on-ai-chip-shipments-applies-to-chinese-firms-outside-china [12] Gibson Dunn, "EU AI Act Omnibus agreement: postponed high-risk deadlines," June 2026. https://www.gibsondunn.com/eu-ai-act-omnibus-agreement-postponed-high-risk-deadlines-and-other-key-changes/ [13] UNESCO, "Global Dialogue on AI Governance, Geneva, 6-7 July," July 2026. https://www.unesco.org/en/articles/global-dialogue-ai-governance-geneva-6-7-july [14] Konstantinos Komaitis, "What the AI Impact Summit reveals," DFRLab, 4 March 2026. https://dfrlab.org/2026/03/04/what-the-ai-impact-summit-reveals/ [15] Syed Akbaruddin, "Beijing's WAICO could determine new global AI order. India must be vigilant," The Indian Express, 11 November 2025 (quote via VisionIAS summary). https://visionias.in/current-affairs/upsc-daily-news-summary/article/2025-11-11/the-indian-express/international-relations/beijings-waico-could-determine-new-global-ai-order-india-must-be-vigilant [16] Caixin Global, "China launches Shanghai-based AI governance body with 29 founding nations," 17 July 2026. https://www.caixinglobal.com/2026-07-17/china-launches-shanghai-based-ai-governance-body-with-29-founding-nations-102465524.html