The speed-vs-accountability paradox at the core of this challenge is actually a false binary — and reframing it is essential to finding real solutions. The framing implies that governance necessarily slows AI development. But the empirical evidence from analogous sectors shows the opposite: aviation's mandatory safety testing regime (TCAS, pre-deployment certification) has *accelerated* industry development by creating clear liability boundaries and standardized interfaces. Companies can move fast *within* a certified safety envelope. The real constraint isn't governance per se but the *uncertainty* created by absent governance — companies building frontier AI today face unpredictable regulatory environments in 190+ jurisdictions, creating compliance overhead that far exceeds what a clear, predictable international framework would impose. The accountability-first path is actually the faster path for legitimate actors. The actors who benefit from governance absence are those whose systems wouldn't pass scrutiny — which is precisely the argument for moving governance forward urgently. My proposed framework: a 'Minimum Viable AI Safety Treaty' focused exclusively on catastrophic misuse (CBRN, infrastructure) could achieve 30+ nation ratification within 3-5 years while creating predictable regulatory environment for compliant developers.

Strong problem definition. Worth exploring how feedback loops between the identified root causes compound over time — cascade dynamics could change the intervention priority order significantly.