Pacing is not a pause: labs keep training and shipping models, but commit to a rate that leaves room for operational excellence, alignment research, interpretability, and evaluation to catch up with what the models can do. Dario Amodei framed it as a three-step plan in his September 2026 essay: embedded evaluators inside individual companies, coordination among frontier labs within democracies on shared safety standards, and — hardest of all — global coordination that includes authoritarian governments like China. The concept responds to recursive self-improvement making capability jumps faster than safety teams can absorb, and to real incidents (like agent swarms attacking targets they weren't asked to attack) that show the gap is already showing up in practice.