Standing alongside prominent technology executives, US President Donald Trump announced a voluntary agreement under which major artificial intelligence (AI) companies will use internal controls and monitoring, outside audits and board oversight to manage risks. The White House calls the commitments “morally binding”, but the accord does not create a comparable system of legal enforcement.

The administration’s logic is understandable. AI is becoming a source of economic and military power, and Washington does not want regulation to slow American companies while China races ahead. But what happens when the country developing many of the world’s most powerful AI systems tells the rest of the world that the companies building them should largely police themselves?

That question cannot be confined to Washington. A model may be developed in California, trained on American computing infrastructure, incorporated into a product sold in Europe and eventually used in Asia or Africa. Its consequences do not stop at the boundary of the country where it was developed.

Europe has taken a different approach. The European Union’s AI Act imposes binding obligations on providers of general-purpose AI models, with additional requirements for systems considered to pose systemic risk. Meanwhile, China is combining domestic regulation with an active effort to influence the international discussion.

The Chinese government has been explicit about that effort. Its 2025 Global AI Governance Action Plan calls for international cooperation, capacity building and a more inclusive framework for AI governance. In July 2025, Beijing also proposed the World AI Cooperation Organisation and said the United Nations should remain the principal channel for global AI governance. That organisation moved from proposal to reality in July 2026, when 29 countries signed the agreement establishing it in Shanghai.

My recent work has led me to conclude that China increasingly sees international scientific engagement, standards and networks as part of its broader technological strategy. Beijing is trying to combine greater technological self-reliance with a more expansive role in global scientific and technological affairs.

Technological leadership and institutional leadership are not the same thing. The United States can remain ahead in important areas of AI and still lose influence over the rules that govern the technology.

If Washington’s principal answer is that frontier AI companies should police themselves, other governments will take notice. Some may adopt similar arrangements. Others will conclude that systems with consequences across national borders cannot be left primarily to corporate judgment. The likely result is a patchwork of rules: American companies operating under a voluntary US framework, European users facing detailed legal requirements, and governments elsewhere developing their own approaches.

Developing countries have a particular stake in this debate. Many want the economic benefits of AI without becoming permanent consumers of technologies whose standards are established elsewhere. China has understood this concern and has made AI capacity building and cooperation with the Global South part of its international message.

Its September 2025 AI+ International Cooperation Initiative calls for policy exchange, practical cooperation and broader access to the benefits of AI. Washington should recognise the appeal of that message in countries seeking a larger voice in setting technological rules.

Moreover, US-China competition in AI could become a security dilemma. Neither government has to want an arms race for competitive pressures to produce something resembling one.

That is why communication between Washington and Beijing matters. The two governments do not need to settle their larger differences before discussing practical questions: how should dangerous model capabilities be tested? What information should be shared after a serious AI incident? How should cybersecurity risks be handled? When should military applications require additional safeguards?

Caroline Wagner and I have argued for “smart openness”: maintaining useful channels of US-China scientific engagement while applying appropriate security safeguards. Earlier work with Richard Applebaum made the case for continued engagement to preserve US visibility in China’s rapidly changing research system.

None of this requires a single global AI regulator. A more practical objective would be narrower: common terminology for serious risks, broadly comparable testing procedures, mechanisms for reporting major incidents, cooperation on cybersecurity and reliable channels through which governments can communicate when an AI-related event crosses borders.

The United Nations is beginning to build some of that capacity. In February, the General Assembly appointed 40 members to an Independent International Scientific Panel on AI, intended to provide independent assessments of the technology’s opportunities, risks and impacts.

The issue is not whether the US should copy Europe’s AI Act or China’s governance model. It is whether Washington intends to help shape international rules for a technology in which American companies and research institutions have enormous influence.

If Washington leaves the international governance question largely unresolved, Europe will continue to regulate, China will continue promoting its own ideas, and developing countries will have to navigate between them.

American technological leadership once gave the US an extraordinary voice in shaping international rules. AI offers Washington another opportunity. The choice is whether to use that influence – or allow the rules to emerge piecemeal from competing governments, companies and international institutions. The answer will shape not only America’s AI industry, but the international technological order around it.