Why the US, China and EU may doom global AI governance, and how to fix it

Artificial intelligence has outpaced every regulatory system built to contain it, and the world’s three most powerful technology blocs cannot agree on who should set the rules. A new study published in the journal Global Public Policy and Governance argues that this deadlock is not merely a diplomatic inconvenience but a textbook collective action problem, one that could be solved with tools developed decades ago to manage fisheries, forests and water supplies. The research, conducted by Olajide Olugbade of the Georgia Institute of Technology’s School of Public Policy, examines why the United States, China and the European Union, despite all recognizing the need for common international standards on AI, remain unable to cooperate, and what lessons from the science of cooperation might break the impasse.

The starting point of the analysis is the fragmented state of AI governance itself. Scholars describe the current landscape as underdeveloped, unorganized and immature, populated by three broad categories of actors. Private sector giants such as Google, OpenAI, Microsoft, IBM and Meta have produced industry guidelines, ethics documents and voluntary commitments, often favoring self-regulation that does not slow the pace of development. Public sector actors, including national governments and intergovernmental bodies like the G20, the Organisation for Economic Co-operation and Development and the United Nations, have published national strategies and international agreements. Non-governmental organizations, professional bodies such as the IEEE, think tanks and advocacy groups round out the picture, setting standards and drawing attention to emerging risks. Yet despite this flurry of activity, there is still no generally recognized global governance mechanism for AI, and no consensus on what one should even look like.

The central obstacle, according to the study, lies in the divergent approaches of the three actors that lead AI development worldwide. The United States has published more AI strategies and policy reports than any other country and pursues a market-driven, less restrictive model, prioritizing sector-specific governance, granting autonomy to federal agencies and allowing industry to shape policy through consultation and voluntary commitments. There is no federal AI law; the most prominent federal action has been the Executive Order on Safe, Secure, and Trustworthy AI, while state legislatures pass a patchwork of regulations on applications ranging from elections to healthcare. China, by contrast, treats AI as an extension of its centrally planned political system, aligning development with national and socialist values through a rule-based approach while tolerating decentralized innovation at the local level. Its 2017 New Generation Artificial Intelligence Development Plan declares the ambition to become the world leader in AI by 2030. The European Union has carved out a third position, weaving the narrative of trustworthy, human-centered AI into global discourse and releasing the EU AI Act, the world’s first comprehensive AI law, which classifies systems into four risk categories with escalating regulatory requirements.

These three actors each want to lead globally, and their competing ambitions generate concrete barriers to cooperation. The EU’s risk-based regulation is perceived by the US as innovation-stifling, while the US preference for self-regulation clashes with both the EU’s legal approach and China’s state-led model. Geopolitics deepens the divide: decoupling technology supply chains, the ongoing chip war and trade tensions have made faster AI innovation a matter of national security for Washington and Beijing, while the EU pursues open strategic autonomy. Ideological differences compound the problem. The US and EU favor collaboration with democratic partners, whereas China’s use of AI for mass surveillance and social scoring has alienated Western states, pushing it toward partnerships with Russia and developing countries through the Belt and Road Initiative and an AI study group within BRICS. Even the preferred modes of cooperation differ, with the US and EU working through existing institutions while China builds new alternative ones, a dynamic that risks accelerating bloc formation in an already fragmented landscape.

The consequences of this non-cooperation are far from abstract. The study catalogs a series of risks that are amplified or directly generated by the standoff between the three powers. Heightened nationalism in AI development could divert attention from global concerns such as sustainability toward short-term economic interests, fostering protectionist policies that disrupt markets and widen inequalities between nations. The US-China rivalry could spill over into other international conflicts as more countries pursue AI purely for competitive advantage, concentrating power among leading nations while dispersing costs widely. A hollowed-out multilateral system could fragment into minilateral groupings, enabling forum shopping in which countries align with the least restrictive regime. Cross-border risks, from AI-driven cyberattacks to nefarious uses of open-source models, become harder to manage without joint research and coordination, and societal-scale risks from frontier AI models could go unassessed without an international consortium pooling expertise and regulatory resources.

Perhaps most alarming is the prospect of a race to the bottom, in which each actor prioritizes speed over safety, cutting corners to be first to achieve breakthroughs. Coordination would help establish AI safety standards and agree on red lines that should not be crossed, including military applications such as lethal autonomous weapons systems and bioweapons. A competitive race toward artificial general intelligence among the three most capable actors could expose humanity to existential and global catastrophic risks, the study warns, making collective management of long-term dangers as urgent as addressing immediate harms.

To diagnose the problem, Olugbade turns to the collective action literature, a field transformed by Mancur Olson’s 1965 book The Logic of Collective Action and later by the Nobel laureate Elinor Ostrom. Collective action problems arise when group members pursue short-term individual interests at the expense of long-term shared interests, leaving everyone worse off than if they had cooperated. The key to classifying such problems lies in the nature of the goods at stake, defined by two properties: excludability, whether access can be limited, and rivalry, whether use by one diminishes availability for others. Global AI governance, the study argues, is non-excludable, since any of the three actors can currently influence AI’s direction, but rivalrous, because dominance by one precludes equivalent influence by the others. That combination makes global AI governance a common good, placing it in the same analytical category as the common-pool resources Ostrom famously studied, from fisheries to irrigation systems.

Ostrom distilled eight design principles from long-enduring institutions that successfully managed common-pool resources, and the study applies each to AI governance. Clearly defined boundaries would mean identifying who has a right to do what, using multistakeholder mechanisms while granting the three leading powers special privileges of influence as incentives to participate. Rules must be congruent with local conditions, since centrally determined universal rules are unlikely to survive the ideological differences among the three actors; regional and local mechanisms must complement global ones. Collective-choice arrangements should allow stakeholders at every level to modify rules as the technology evolves. Monitoring and graduated sanctions are deemed crucial, addressing the enforcement weakness that has undermined initiatives like the UNESCO Recommendation on the Ethics of AI and the OECD AI Principles. Mutual monitoring between the US and China, each interested in the other’s activities, could be leveraged, with sanctions escalating progressively based on compliance history. Conflict-resolution mechanisms, minimal recognition of the rights of governance bodies to organize, and nested enterprises complete the framework.

The study also draws on the analytical framework for large-scale collective action developed by Sverker Jagers and colleagues, which identifies facilitators such as trust, reciprocity and reputational stake, and stressors such as anonymity, lack of accountability, heterogeneity, risk and uncertainty. Applied to AI governance, the facilitators include a shared perspective among all three actors on using AI to solve global challenges and a common willingness to work with international organizations, with China engaging the UN and the US and EU supporting the OECD. These glimmers of common ground, the study suggests, could serve as entry points around which global AI governance is organized, with international organizations such as the UN, the OECD and the Global Partnership on AI acting as third-party interventions with the capacity and legitimacy to generate cooperation, including through the newly proposed International Scientific Panel on AI.

The study’s central conclusion challenges the many proposals for a single centralized global AI authority, such as a G20 coordinating committee, an International Artificial Intelligence Organization or an Intergovernmental Panel on AI modeled on the climate body. Such centralized mechanisms, the analysis finds, are less likely to succeed than a polycentric, multilevel arrangement in which interconnected but independent governance institutions operate at global, regional, national and subnational scales, each determined by the stakeholders at that level and guided by the principle of subsidiarity. Polycentric systems allow learning from experimentation, build trust, respond to local contexts and reduce opportunistic behavior. Enforcement, supported by monitoring, sanctioning, information provision and conflict resolution, remains the linchpin of implementation. The author acknowledges that the paper focuses on institutional design considerations rather than specifying exact institutions, and calls for future research on how polycentric arrangements could be implemented and how information provision should be structured. But the message is clear: any global AI governance proposal that ignores the non-cooperation of the United States, China and the European Union is likely doomed to fail, and the path forward runs not through a single world AI authority but through a resilient mosaic of overlapping, mutually adjusting institutions.

Subject of Research: Global governance of artificial intelligence analyzed through collective action theory

Article Title: In search of a global governance mechanism for Artificial Intelligence (AI): a collective action perspective

Article References: Olugbade, O. (2025). In search of a global governance mechanism for Artificial Intelligence (AI): a collective action perspective. Global Public Policy and Governance, 5(2), 139-161. https://doi.org/10.1007/s43508-025-00113-z

Image Credits: AI Generated

DOI: 10.1007/s43508-025-00113-z

Keywords: artificial intelligence, global governance, collective action, common-pool resources, Elinor Ostrom, polycentric governance, EU AI Act, US-China rivalry, institutional design, AI regulation, international organizations, frontier AI risks