Adapting Lessons from the Nuclear Non-Proliferation Era for the Global Governance of Artificial Intelligence

Science & Technology 2025-2026 High School Essay Contest Winner

Artificial intelligence (AI) is rapidly transforming global security, economic competition, and political power, creating governance challenges comparable to those posed by nuclear weapons after the Second World War. Although AI differs fundamentally from nuclear technology in its accessibility, speed of development, and reliance on private industry, both technologies present existential risks that no state can effectively manage alone. This article argues that the international community should adapt key principles from the postwar nuclear governance system, particularly the Nuclear Non-Proliferation Treaty (NPT) and the International Atomic Energy Agency (IAEA), to create an international framework for AI governance. Rather than focusing exclusively on software, such a framework should emphasize verification of advanced computing hardware, establish an International Artificial Intelligence Agency (IAIA), and create a global grand bargain that provides developing nations with access to AI resources in exchange for compliance with internationally agreed-upon safety standards. Ultimately, long-term global security in the AI era will depend less on technological secrecy than on transparency, verification, and collective restraint.

Introduction

The mid-twentieth century was defined by the “Old Order” of atomic diplomacy, a period where our survival hinged on the management of a singular, tangible, and state-controlled technology: the split atom. Today, we are at the cusp of a second great transition, one in which the “New Order” is being forged not in laboratories of enriched uranium, but in the server farms of Silicon Valley and Shenzhen. As AI begins to reshape statecraft, warfare, and economic productivity, the international community faces a familiar anxiety. Nonetheless, while the threats of AI mirror the existential risks of the nuclear age, our current governance models are lagging behind. To navigate this transition, we must focus on the lessons that states and movements can learn from the “Old Order,” specifically the Nuclear Non-Proliferation Treaty (NPT) and the International Atomic Energy Agency (IAEA), while critically adapting them to the decentralized, intangible nature of the digital age. 

The parallel between nuclear energy and AI lies in their “dual-use” nature. In the 1940s and 50s, the world struggled to balance the benefits of nuclear power with the horror of the hydrogen bomb. A similar dichotomy arises with AI, which promises to solve complex climate modeling and cure diseases, yet also enables autonomous weapons systems and the danger of deepfakes. As in the “Older Order,” the primary driver of this international tension is the “Security Dilemma.” Similar to how the United States and the Soviet Union felt compelled to expand their arsenals to ensure mutual assured destruction, modern states fear that a unilateral pause in AI development will result in a permanent strategic disadvantage. The “Old Order” managed this through formal treaties and physical verification. The “New Order” requires a similar commitment to transparency, yet actors have changed. Power has shifted from a state monopoly to a handful of private corporations, necessitating a transition in how we define “sovereign responsibility.” We are no longer merely managing a race between Washington and Moscow, but rather a complex, multipolar race in which the “labs” are often more powerful than the regulators. 

The nuclear order was defined by scarcity and visibility, as for states to build a bomb, they require massive industrial infrastructure, rare minerals, and specialized facilities that could be tracked via satellite. Consequently, the IAEA’s safeguards relied on physical inspections of reactors and centrifuges. This was a “top-down” order where the barriers to entry were so high that only a few actors could participate. Yet for AI, the most necessary equipment for advanced AI is code and data, which are non-physical and can be transmitted across borders in milliseconds. While hardware, specifically high-end semiconductors (GPUs), remains a physical bottleneck, the democratization of open-source models means that non-state actors can wield significant power without the industrial footprint of a nation-state. This creates a vertical transition in power where power moves downward from the state to the individual and outward from the government to the private firm. Our current trajectory seems to be attempting to apply a “Westphalian” solution to a “Post-Westphalian” technology. Current AI safety discussions, such as those found in the Bletchley Declaration, often focus on voluntary commitments from major tech firms. However, history shows that voluntary compliance is the hallmark of a failing order. In 1946, the Baruch Plan proposed the internationalization of nuclear energy, but it failed since it lacked a mechanism for enforcement that all parties trusted. We are currently repeating the Baruch Plan’s failure by relying on meetings such as AI Safety Summits that produce vague communiqués rather than binding technical protocols. Without a global framework that offers emerging economies access to “AI for Good” in exchange for safety compliance, the world risks a fractured “Digital Iron Curtain.” 

A unique feature of this transition is the reshaping of the state itself. In the nuclear era, the private sector was a contractor for the government. In the AI era, the private sector is the innovator and owner of the technology. Companies like OpenAI or DeepMind currently hold more sovereign knowledge about the capabilities of Frontier Models than most intelligence agencies. This creates a crisis of legitimacy. If the “New Order” is governed only by corporate boards whose primary fiduciary duty is to shareholders rather than global stability, the social contract is fundamentally broken. Therefore, through our current governments, we must integrate these private actors into the treaty framework. A successful example of this would be the Chemical Weapons Convention (CWC), which requires private industrial facilities to submit to international inspections, as a bridge. A transition to a stable AI order requires that we stop treating tech companies as mere vendors and start treating them as high-risk entities subject to the same international oversight as a nuclear enrichment plant.

The transition to a new AI order also reshapes the relationship between the state and its citizens. In the nuclear age, the public’s role was largely one of protest, characterized by movements like the Campaign for Nuclear Disarmament. In the AI era, civic participation is more direct because the technology is integrated into daily life. Citizens are not just victims of potential AI conflict; they are the providers of data and train these systems. Legitimacy in the New Order will not come from state secrecy, but from algorithmic accountability. If governments use AI for surveillance or judicial decisions, they must adopt the same “verify, then trust” mindset that defined Cold War arms control. The reshaping of civic participation means that international governance can no longer be a closed-door meeting of diplomats, but should include technical experts, civil society, and the private sector as equal stakeholders. This transition represents a shift in how political legitimacy is constructed, where states provide physical security through the ability to provide digital security and data privacy. 

Implementing the Three Policy Pillars

To successfully navigate this transition, the international community should implement three policy pillars, learned from the successes and failures of the nuclear era. Firstly, governments should create an International Artificial Intelligence Agency (IAIA): Modeled after the IAEA, this body would not act as a global dictator of code, but as a center for technical standards and verification. Its primary goal would be to establish “safety guardrails” for Frontier Models. Similar to how the IAEA monitors the diversion of nuclear material, the IAIA would monitor the “diversion” of massive compute clusters toward malicious uses, such as bio-weapon design or autonomous offensive cyber-capabilities. This proposal aligns with recommendations from the UN High-Level Advisory Body on AI. 

Secondly, we must implement compute-based non-proliferation. Since code is difficult to track, the “New Order” must focus on the physical bottleneck, which is specialized hardware. States should cooperate on a Hardware Safeguards regime. By embedding “on-device” verification in the next generation of high-end AI chips, the international community can ensure that massive compute power is being used for declared, peaceful purposes, like how the NPT tracks fuel rods in a civilian power plant. This allows for verification without infringing on the proprietary intellectual property of the software, solving the tension between trade secrets and global safety. 

Third, there needs to be a “grand bargain” for the Global South. To prevent a “Digital Divide” that mirrors the “Nuclear Haves and Have-Nots,” an international AI framework must include a technology transfer clause. Developed nations and private firms should provide subsidized access to compute and pre-trained models for sustainable development—such as precision agriculture and disaster response—to any state that agrees to strict, verifiable AI safety and human rights standards. This mimics Article IV of the NPT, which guarantees the right to peaceful nuclear energy in exchange for non-weaponization. This “carrot and stick” approach is essential for gaining the global consensus required for a global order. 

The transition from the “Old Order” of nuclear physics to the “New Order” of digital intelligence is the defining challenge of our generation. History teaches us that technology-driven shifts in power are often accompanied by conflict unless the international community proactively designs systems of mutual restraint and cooperation. By adopting a comparative lens, we can see that the age of AI differs from the atomic age of the 1940s. States must learn that in an interconnected era, absolute digital sovereignty is an illusion. Security is found not in the secrecy of one’s own algorithms, but in the collective reliability of the global network. If we fail to learn the lessons of the nuclear age—that existential threats require a surrender of some autonomy for the sake of survival—we risk a transition into an era of chaos rather than a new order of progress. The transition is inevitable; what order we build within it is not.

. . .

Luke Mendoza is a junior at Valencia High School.