The departures are the visible symptom of a quieter institutional failure: the race to the top on AI safety is dead, killed not by a single decision but by a sequence of compromises — Anthropic revising its core safety policy, OpenAI taking Pentagon contracts, the international consensus framework losing its enforcement teeth. When the structures that gave safety researchers institutional standing dissolve, the researchers who remain have no mechanism to escalate concerns. Leaving is not a protest — it is the only remaining option.