The development of artificial intelligence poses a novel governance challenge: can transformative technology be controlled before catastrophic failure demonstrates its necessity? From 1946 onwards, policymakers and institutional actors began confronting this challenge with nuclear weapons and explored governance mechanisms for dealing with existential technological risks. I focus on the years 1946 to 1970 and review this period for lessons applicable to contemporary AI regulation and find the following: Comprehensive prohibition schemes can attract broad support when confronting existential risks, but this support proves politically unworkable without verification mechanisms. International cooperation is likely to fragment across competing jurisdictions, and perhaps critically so. Corporate self-regulation may appear adequate during development, both generally and in particular within commercial contexts, but also proves systematically inadequate under deployment pressure. Verification mechanisms may play a decisive role through technical inspection regimes, but these require decades to establish. The integration of AI into nuclear command systems is likely to create unregulated convergence risks, yet no international framework addresses this. Overall, governance may look more like responding to near-catastrophe than proactive risk management. The IAEA inspection regime demonstrated that monitoring succeeds where the Baruch Plan's prohibition failed. Progress came incrementally through agreements accepting imperfection. Catastrophic proximity drove political will that diplomacy alone could not. Effective regulation may be achievable, but there are substantial obstacles to implementing it before disaster compels action.
Meriel Batterley (Sat,) studied this question.