The Non-Proliferation Analogy

In the AI Story

The NPT, signed in 1968 and in force since 1970, operates through a grand bargain: nuclear-armed states commit to disarmament while non-nuclear states commit to non-acquisition, with all parties obtaining access to peaceful nuclear technology. The treaty's success has been partial — nine nations have acquired nuclear weapons despite the treaty, and the disarmament commitments of the original nuclear powers have been largely unmet — but the structure has proven remarkably durable. One hundred ninety-one nations are parties. The normative framework the treaty established has constrained nuclear proliferation below what most analysts predicted in the 1960s.

Nye's interest in the NPT as analogy derives from its mechanism rather than its specifics. The treaty succeeded because it offered something valuable to non-nuclear states (peaceful nuclear technology, constraint on nuclear weapon expansion) in exchange for their non-acquisition commitments. Nuclear powers did not compel compliance; they attracted it by making the treaty's bargain attractive enough to sign. This is precisely the soft power mechanism: voluntary alignment produced by genuine mutual benefit rather than coercion. An AI governance framework of comparable legitimacy would need to offer similar mutual benefit — protection from AI harms, access to AI benefits, participation in governance — in exchange for commitments to specific constraints on AI development and deployment.

The AI governance challenge differs from the nuclear challenge in important respects. AI capability diffuses faster than nuclear capability. AI development occurs primarily in the private sector rather than through state programs. AI has dual-use characteristics more pervasive than nuclear technology. These differences complicate the analogy but do not invalidate it. The underlying mechanism — legitimate governance framework attracting voluntary compliance through genuine mutual benefit — remains the template. What differs is the specific institutional architecture required to implement the mechanism for AI's particular characteristics.

The analogy also illuminates what AI governance must avoid. Governance frameworks perceived as serving the narrow interests of AI-advanced nations rather than humanity broadly will generate resistance proportional to the asymmetry. Frameworks that fail to offer meaningful benefits to non-leading nations will not attract voluntary compliance. Frameworks that attempt to freeze current AI distributions rather than enabling equitable development will be seen as digital colonialism. The nuclear precedent suggests that the architecture requires careful construction: the bargain must be genuine, the benefits must be real, the obligations must be distributed, and the institutional machinery must be perceived as legitimate across the participating nations. Building such an architecture for AI is the defining smart power challenge of the era.

Origin

Nye invoked the analogy repeatedly across his AI commentary, most systematically in his 2024 columns on AI and national security. The framework draws on his own experience as Deputy Under Secretary of State for Security Assistance, Science, and Technology (1977-1979) in the Carter administration, where he participated directly in non-proliferation policy implementation.

Key Ideas

Legitimacy over coercion. The NPT succeeded because nations voluntarily joined, not because they were compelled; AI governance must operate through the same mechanism.

Genuine mutual benefit. Voluntary alignment requires bargains that genuinely serve participating nations' interests, not arrangements that serve the narrow interests of leading powers.

Institutional architecture. Legitimate frameworks require specific institutional machinery — treaty bodies, verification mechanisms, dispute resolution — that takes time to build and requires sustained investment.

Dissimilarities matter. AI's faster diffusion, private-sector development, and pervasive dual-use characteristics complicate but do not invalidate the analogy's core mechanism.

Constructive risks. Governance frameworks perceived as entrenching existing asymmetries rather than enabling equitable development generate resistance that undermines the soft power legitimacy they seek to project.

Debates & Critiques

Critics question whether the NPT model is applicable given AI's distinctive characteristics — particularly its rapid diffusion and private-sector locus. Some argue that AI requires fundamentally different governance architecture. Others argue that the NPT itself has been less successful than Nye's analogy suggests. The book treats these as refinements rather than refutations: the specific institutional machinery must be adapted, but the underlying mechanism of legitimate framework attracting voluntary compliance remains the most viable model for a governance challenge that cannot be solved through coercion.