Key Points
- International coordination on AI is necessary because data, AI models, and cyber risks cross borders. Yet governments are producing competing frameworks rather than a shared governance regime.
- Disagreement over what AI governance should achieve runs through the democracies as well as between the United States and China. New coalitions such as Pax Silica and WAICO prioritize supply chains, investment, and access to AI technology over managing AI-related security risks.
- Fragmentation raises compliance costs for firms, narrows the options available to smaller states, and shifts important decisions—from access to advanced chips to the management of AI safety incidents—toward narrower groups of powerful actors.
On July 16, 2026, representatives of 29 countries gathered in Shanghai to sign the founding agreement of the World Artificial Intelligence Cooperation Organization (WAICO), a new intergovernmental body headquartered in the city. Its founding members included China, Russia, Kazakhstan, Pakistan, and Indonesia, alongside countries from across Asia, Africa, Latin America, and Eastern Europe; eight more states joined later that month. No G7 country was among the founding signatories. By then, Japan, the United Kingdom, and the United States had already signed Pax Silica, a U.S.-led initiative launched in December 2025 to secure supply chains for semiconductors, critical minerals, and AI infrastructure; the European Commission followed on behalf of the EU in June 2026. The emerging divide soon became more explicit. In August, the U.S. State Department was reported to be drafting a letter to countries that had signed an AI Opportunity Statement linked to Pax Silica, warning that participation in the U.S.-led initiative would be incompatible with joining WAICO. Bangladesh, meanwhile, announced it would join WAICO as an observer on the same day a U.S. envoy raised the prospect of a Pax Silica partnership.
Rather than converging on a common set of rules for AI, governments are developing overlapping frameworks and coalitions, and a growing number of countries are being asked to navigate among them. Why does AI require international coordination, and why are these efforts producing competing arrangements rather than convergence? This article argues that the answer lies partly in a deeper disagreement over what AI governance is for, and partly in who holds the power to shape it. The newest coalitions place greater emphasis on supply chains, investment, and access to AI technology than on managing cross-border AI-related risks. The costs of this fragmentation fall disproportionately on smaller and developing countries, which have less influence over how these arrangements evolve.
A Problem No Country Can Govern Alone
The first article in this series identified three risks where AI meets economic security: data leakage, algorithmic bias and information control, and AI-enabled offensive capabilities. None stops at national borders. Data entered into an AI service in one country may be stored or accessed in another, where different laws apply. An AI model developed under one country’s legal and political constraints can shape the information presented to users elsewhere. AI-enabled cyberattacks can cross borders with little regard for national jurisdiction. The underlying infrastructure is equally transnational. AI development depends on chip supply chains that no single country controls and on data that flows across borders. It is no accident that trade agreements, the main avenue so far for trade-related AI cooperation, increasingly address AI alongside the data flows on which it depends.
National governance is therefore necessary but insufficient. A government can regulate the firms within its jurisdiction, but not the foreign data centers, AI models, and suppliers on which those firms depend. A company operating under one country’s rules may also become subject to another country’s legal requirements as its operations cross borders. The case for international coordination is strong. Yet coordination is proving elusive, and the obstacles are often political rather than technical. Governments disagree not only about how to govern AI, but about what AI governance is for.
Where the Disagreements Lie
Three fault lines structure the international debate. Each reflects, in different ways, a tension introduced in the first article: the openness that drives innovation also creates vulnerability, while the restrictions that reduce risk can undermine competitiveness. At the international level, an additional question arises: who gets to manage those trade-offs, and whose interests count?
The first fault line runs between the United States and China, which are competing to set the terms on which other countries adopt AI. Washington’s current approach pairs domestic deregulation with trusted-partner supply chains. At the February 2026 AI Impact Summit in New Delhi, the head of the U.S. delegation declared that the United States “totally” rejects global governance of AI, and Washington declined to back the second International AI Safety Report, an expert assessment of risks from advanced AI. The AI Opportunity Statement signed at the Pax Silica Summit in June similarly promotes a pro-growth, pro-innovation approach to regulation. Beijing, by contrast, emphasizes sovereignty, development, and broader access to AI technology. At the conference following WAICO’s launch, Xi Jinping pledged 5,000 AI training opportunities for developing countries over five years. Despite their differences, what each side offers other countries emphasizes growth, access, and technological capacity over managing AI-related security risks.
The second fault line runs through the democracies themselves. The EU remains the reference point for binding regulation, yet even it is recalibrating under pressure to stay competitive. In May 2026, EU legislators agreed to postpone some of the AI Act’s obligations for high-risk uses, such as AI in hiring and credit decisions, by more than a year. Japan and the United Kingdom have relied mainly on guidance and voluntary compliance, while the United States has moved further toward deregulation, with the White House pushing for a national framework that would override many state-level AI regulations. Democracies also disagree over how much international oversight they are willing to accept. On September 21, twenty countries and the EU called for governments to explore the creation of an international institution to set standards and verify compliance. Germany, Canada, and Australia signed; Japan, the United Kingdom, France, and South Korea did not. Even close partners do not share a view of how far international oversight should go.
The third fault line divides North and South. Many developing countries have had limited influence over the rules and institutions shaping AI governance, despite significant exposure to its risks and a substantial stake in its benefits. WAICO seeks to address this gap through training and capacity-building. The international AI summit series has also sought broader participation. The New Delhi Declaration drew 91 endorsements, including from both the United States and China, but researchers note that it could win such wide support because the declaration was broad in scope and imposed no binding commitments. As geopolitical rivalry, regulatory divergence, and development concerns become intertwined, convergence becomes more difficult.
Many Forums, Thin Commitments
These disagreements play out across a growing number of forums, exposing a familiar challenge in international cooperation: how to combine broad participation with meaningful commitments.
Broad forums tend to produce thin commitments. The UN established a Global Dialogue on AI Governance in August 2025, giving every member state a seat, but its sessions consist mainly of high-level national statements, with no current plans to launch negotiations. The international AI summit series has grown from 29 signatories at Bletchley Park in 2023 to more than 90 in New Delhi in 2026. Along the way, its focus has shifted from the safety of advanced AI toward investment and adoption, with less emphasis on specific governance commitments.
More focused initiatives face their own limits. Japan’s G7 Hiroshima AI Process developed broad guiding principles for AI actors and a voluntary code of conduct aimed at developers of advanced AI systems, including technology companies and research institutions. Its Friends Group has since expanded beyond the G7 to include countries across the Global South, but wider participation has not changed the voluntary nature of its commitments. The OECD AI Principles have served as a broader and longer-standing reference point for trustworthy AI, informing national policies and governance frameworks across many countries, yet they too remain non-binding. The International Network of AI Safety Institutes, since renamed the International Network for Advanced AI Measurement, Evaluation and Science, offers a different model: deeper technical cooperation on AI testing and evaluation, but among a smaller group of countries, most of them advanced economies.
AI governance also takes place outside dedicated AI forums. The EU AI Act applies to companies offering AI systems in the EU, including those based outside Europe. This gives Brussels potential reach beyond its borders through what legal scholar Anu Bradford has called the “Brussels effect,” though how far that effect will extend to AI is debated. International standards bodies such as the International Organization for Standardization (ISO) and the International Electrotechnical Commission (IEC) publish standards for how organizations manage and audit AI, often outside the immediate view of policymakers. Trade agreements, some of which already include rules on cross-border data, provide another venue. The Digital Economy Partnership Agreement, founded by Singapore, Chile, and New Zealand, addresses AI alongside cross-border data flows.
The newest coalitions add another layer. Pax Silica and WAICO have expanded quickly, and, as noted above, neither is designed primarily to constrain AI itself. WAICO also illustrates how institutional boundaries can become blurred. UN Secretary-General António Guterres attended WAICO’s founding ceremony, and Indonesian officials have described the organization as operating under the UN framework, even though its founding agreement establishes it as an independent intergovernmental organization headquartered in Shanghai. The result is a crowded field of overlapping initiatives, with different participants, purposes, and levels of commitment, but no broadly agreed set of binding international AI rules.
What Fragmentation Costs
For firms, the costs are practical. A company operating across jurisdictions must navigate different regulatory requirements, keep up with changing implementation timelines, and manage restrictions on cross-border data. Different rules can also encourage companies to develop or deploy AI in countries where requirements are less strict, allowing risks to migrate along with them.
For governments, fragmentation can concentrate decision-making in fewer hands. When broad forums cannot agree, important decisions may shift to narrower settings: a bilateral channel, a supplier coalition, or a single market with sufficient regulatory reach. For smaller and developing countries, this can mean pressure to choose between competing arrangements at the cost of future options. Kazakhstan joined Pax Silica in June and WAICO in July. Several other WAICO members, including Malaysia, Serbia, Vietnam, and Georgia, have signed the U.S.-led AI Opportunity Statement. Such hedging remains possible while commitments are loose. A U.S. demand for exclusivity, particularly if tied to access to advanced chips, could narrow that space quickly.
Japan faces a different version of this pressure. As a founding participant in Pax Silica and a signatory to the AI Opportunity Statement, Japan is closely aligned with U.S.-led technology initiatives. Pax Silica’s emphasis on trusted supply chains and cooperation among like-minded partners reflects a more selective approach than the broader international cooperation Japan has sought to advance through the Hiroshima AI Process. These approaches can coexist: cooperation on technology security need not preclude cooperation on shared AI governance principles. The challenge arises if competition between U.S.- and China-led initiatives begins to discourage countries from participating in both. Japan could then find it harder to engage WAICO participants—particularly those seeking to maintain ties with both sides—through the Hiroshima AI Process.
Fragmentation need not be permanent. Analysts have argued that durable governance regimes are more often forged after systemic failure than through foresight, and debates on AI frequently invoke precedents from other fields. For example, the Montreal Protocol, agreed in 1987 after scientists showed that certain man-made substances were depleting the ozone layer, became the first treaty in UN history to achieve universal ratification, helped by a fund that finances developing countries’ compliance. The Financial Stability Board emerged after the 2008 crisis exposed the costs of inadequate international coordination and works largely through non-binding standards. Both regimes followed a point at which the costs of inaction had become concrete and widely shared. AI governance may lack those conditions for now: the evidence on its most serious risks is still developing, and there is no jointly governed mechanism that brings developing countries in on shared terms. As the costs of fragmentation become more visible, however, the political incentives for coordination may change.
Who Writes the Rules
Late September captured this fragmented landscape in miniature. In the same week that the twenty countries and the EU called for stronger multilateral standards and verification, Washington and Beijing agreed to launch a bilateral “Super Intelligence (SI) Dialogue” and a communication channel for serious AI incidents. Risk management, in other words, has not disappeared from the agenda. But the bilateral dialogue suggests that some important decisions may increasingly be made outside broader multilateral forums. The question is not simply what rules will govern AI, but who will shape them, in which forums, and on what terms. For Japan, which has invested heavily in multilateral AI norm-setting, the question is how effective that approach can remain in an increasingly fragmented landscape. How Japan is navigating this environment, and where its domestic governance architecture holds and where it strains, is the subject of the final installment.
Photo: U.S. Department of State <Pax Silica Summit> (cropped)
