Nuclear weapons governance took twenty-three years to produce its first binding international agreement. From Hiroshima in 1945 to the Non-Proliferation Treaty of 1968, every attempt at regulating the technology through reasoned argument alone failed: the Baruch Plan collapsed when it asked the Soviet Union to abandon nuclear ambitions on trust, comprehensive prohibition attracted rhetorical support but no enforcement, and voluntary commitments proved inadequate under deployment pressure. What finally generated political will was not diplomacy but proximity to annihilation during the Cuban Missile Crisis of October 1962. This paper argues that artificial intelligence governance is repeating the same pattern. Drawing on the period 1946 to 1970, the analysis identifies what this paper terms the catastrophe-dependency model: effective regulation of existential technology emerges only after near-catastrophic failure demonstrates its necessity. Contemporary AI governance reproduces the structural failures visible in early nuclear regulation: fragmented international coordination, unenforceable voluntary commitments, corporate self-regulation that collapses under commercial pressure, and verification mechanisms that demand decades to build. The most urgent dimension of this pattern is one the nuclear precedent did not face. AI systems are now being integrated into nuclear command and control infrastructure, creating unregulated convergence of two technologies with no international framework governing the integration. The governance gap identified here is not speculative; it exists now, and the historical record suggests crisis has historically preceded meaningful closure.
Meriel Batterley (Sat,) studied this question.