For years, the narrative surrounding artificial intelligence has been framed as a high-stakes geopolitical sprint between two superpowers. In this version of the story, the United States and China are the only players that matter, while the rest of the world is relegated to the role of spectator or customer, waiting to see which ideological operating system wins the day.
But for those tracking the actual movement of policy and code, the reality is far more fragmented. While headlines focus on the “race,” a more quiet and complex system of AI’s shadow global governance is emerging. This isn’t a formal treaty signed in a grand hall, but a patchwork of corporate terms of service, voluntary safety commitments, and overlapping regional regulations that are effectively setting the rules for the digital age before official international law can even catch up.
This shadow governance is characterized by a shift in power from sovereign states to the entities that control the “compute”—the massive clusters of GPUs and data centers required to train frontier models. When a handful of private companies decide which guardrails to implement in their APIs or which countries to restrict from accessing their most powerful models, they are performing a regulatory function that was once the exclusive domain of governments.
The Corporate Architects of Global Norms
The most immediate form of shadow governance happens in the boardrooms of Silicon Valley, and Seattle. Because the technical complexity of large language models (LLMs) far outpaces the expertise of most legislative bodies, governments often rely on the companies themselves to define what constitutes “safe” or “responsible” AI.
This creates a feedback loop where corporate safety standards become the default global benchmark. For instance, the “red-teaming” processes used by developers to find vulnerabilities in their models are now being adopted as standard practice by regulators worldwide. While these voluntary measures are a start, they operate without democratic oversight, leaving the definition of “harm” to be decided by internal corporate ethics boards rather than public debate.
the control of hardware creates a physical layer of governance. The U.S. Government’s use of export controls on high-end chips, such as those produced by NVIDIA, serves as a blunt instrument of policy. By restricting the flow of compute, the U.S. Is not just competing in a race but is actively shaping who is allowed to participate in the development of frontier AI, effectively governing the global capacity for innovation through supply chain management.
A Patchwork of Regulatory Philosophies
While the shadow governance of corporations fills the void, official state efforts have resulted in a disjointed landscape of regulatory philosophies. Rather than a single global standard, we see three distinct poles emerging, each attempting to project its influence globally.
The European Union has taken the lead in codified law with the EU AI Act, which entered into force on August 1, 2024. This framework is risk-based, banning certain “unacceptable” AI uses and imposing strict transparency requirements on high-risk systems. The EU is betting on the “Brussels Effect,” hoping that companies will standardize their global products to meet the EU’s strict requirements rather than maintaining different versions for different markets.
In contrast, the United States has largely leaned toward a decentralized, market-driven approach, punctuated by executive actions. The October 2023 Executive Order on the Safe, Secure, and Trustworthy Development and Use of Artificial Intelligence focuses on safety testing and national security, but it stops short of the comprehensive legislative mandates seen in Europe.
China has pursued a more state-centric model, implementing specific regulations for generative AI and algorithmic recommendations that prioritize social stability and alignment with state values. This creates a world where the “rules of the road” for AI depend entirely on where the server is located or where the user is clicking.
| Region | Primary Mechanism | Core Philosophy | Key Focus |
|---|---|---|---|
| European Union | EU AI Act (Law) | Rights-based / Precautionary | Risk categorization & fundamental rights |
| United States | Executive Orders / Voluntary | Innovation-led / Security | Safety testing & national security |
| China | Targeted Regulations | State-led / Control | Content alignment & social stability |
The Compute Divide and the Global South
The most concerning aspect of this shadow governance is the widening “compute divide.” While the U.S., China, and the EU debate ethics and safety, much of the Global South is facing a new form of digital colonialism. The infrastructure required to build sovereign AI—massive data centers and specialized hardware—is concentrated in a few wealthy nations.
This means that countries in Africa, Southeast Asia, and Latin America are often forced to import AI models that are trained on Western data and reflect Western cultural biases. When these models are integrated into local healthcare, education, or legal systems, they bring with them a hidden layer of governance—embedded assumptions about truth, morality, and law that were never vetted by the people using them.
Efforts to bridge this gap are underway, but they remain fragmented. The United Nations has established an AI Advisory Body to foster international cooperation, but it lacks the enforcement power to challenge the dominance of the compute-rich nations. The result is a world where AI governance is not about universal human rights, but about who owns the chips and the data.
What This Means for the Future
The transition from a “race” narrative to a “governance” narrative reveals that the real struggle is not about who gets to the finish line first, but who gets to write the rulebook. The shadow global governance currently in place—driven by corporate interests and fragmented state laws—is efficient but fragile. It lacks the legitimacy of a global consensus and the flexibility to adapt to the rapid evolution of the technology.
The risk is that we settle into a “splinternet” of AI, where different regions operate under entirely different sets of algorithmic truths and ethical constraints. This would not only hinder scientific collaboration but could exacerbate geopolitical tensions as AI systems become tools for ideological enforcement rather than productivity.
The next critical checkpoint for formalizing these rules will be the upcoming series of international AI Safety Summits and the continued implementation phases of the EU AI Act throughout 2025, which will force companies to disclose more about their training data and energy consumption. These milestones will determine if the shadow governance of the present can be brought into the light of transparent, democratic oversight.
This article is provided for informational purposes only and does not constitute legal or financial advice regarding AI compliance or investment.
Do you believe AI governance should be led by international treaties or left to market innovation? Share your thoughts in the comments or share this piece with your network.
