How — and who — should govern AI?
Today President Trump signed the opening salvo for US AI Regulation with a much anticipated executive order. While mostly voluntary, the order opens the potential for more government review of AI models. But Washington isn’t the only place where AI governance is top of mind — China and Europe are also focused on how to manage this rapidly evolving technology.
The US and China are the two global powers most heavily invested in leading in the development, adoption, and export of AI. Each seeks to lay the foundations of the global AI stack — meaning, all the component parts needed for a competitive advantage in AI from training data, energy abundance, and data center infrastructure, to advanced chip design and production, and human talent — all at mass scale.
Across these components, the US leads but Beijing’s massive investments, as well as rampant theft of intellectual property that has gone on for decades, are starting to addup. The Trump-Xi summit produced a tentative agreement for the US and China to discuss the safe development of AI — a fragile but consequential opening for both sides to shape the future of global AI governance.
What is clear, however, is that governments have limited ability to regulate AI and emerging tech. The typical top-down regulatory approach, whereby governments impose regulation on a specific sector — pharmaceuticals, finance, utilities, etc. — simply cannot keep up with the rapid development of technologies that are revolutionizing every sector simultaneously. Europe exemplifies the shortcomings of the top-down model. The EU has led the way when it comes to tech and AI in particular. In 2024, the EU became the first government body to establish a regulatory framework for AI based primarily on risk mitigation.
Even in the two years since the AI Act, AI capabilities have outgrown the regulation. That is because legislators, who are not technologists, cannot foresee the future development of AI capabilities, from advanced generative AI to agentic AI systems, thus making regulations out of date almost upon arrival. Second, the focus on risk effectively has meant a hobbling of innovation, which has limited European companies’ ability to develop frontier AI models, slowed the adoption of AI in Europe, and discouraged US AI firms from deploying their technology to European customers and consumers. This has left Europe lagging further and further behind in the global AI race.
Europe’s experience highlights the fundamental challenge the traditional top-down governance model poses when it comes to AI: it may have worked with railroads, but it does not with a technology as transformative and all-encompassing as AI. A key reason for this is that a top-down AI governance model sees technology as the problem (e.g.risk), rather than as a solution.
But if governments are ill-equipped to lead on AI governance, then who will? This is one key question driving much of the bitter competition between the most highly valued AI firms. Anthropic, for example, has made a public case for firms demanding and embedding certain safety protocols in their models, especially as their capabilities grow. Other leading companies like OpenAI, maker of ChatGPT, have published their own safety and security frameworks “that go beyond current legal requirements.”
Indeed, newer AI firms are already establishing evaluation mechanisms and embedding constraints, permissions, and safeguards into code, compute infrastructure, and deployment architectures. Technical design choices are shaping levels of openness, accountability, transparency, safety thresholds, and acceptable use of AI. The AI systems themselves — not statutes — are, in turn, shaping how models and algorithms are used across the globe.
Bottom-up governance, embedded in the models themselves, may be the future of tech governance writ large. Ultimately, this means the most consequential tech race may not be for the highest performance frontier model, but over the principles that are embedded in those models. Closing the gap between those who build AI and those who govern it is now a central challenge of the AI age.
I found the below 5 pieces on this debate particularly interesting.
Dr. Alina Polyakova serves as the President and CEO of the Center for European Policy Analysis (CEPA) and is the Donald Marron Senior Fellow at the Henry A. Kissinger Center for Global Affairs at the Johns Hopkins University’s School of Advanced International Studies (SAIS).
