Thomas Schelling
Cited (CIT-008, The Strategy of Conflict 1960; Arms and Influence 1966) for the central insight the corpus borrows: deterrence's difficulty is not the physics, it is the commitment capability of the party making the threat.
In the Corpus
Schelling is the vault's most consequential external import, because he is what turns the bomb analysis from an intuition into an argument. The corpus's game matrix (L536) shows the human's dominant strategy is to detonate unless the human can credibly commit not to — and the AI cannot distinguish "not detonating now" from "not detonating ever". The AI's expected value of cooperation is therefore bounded above by the human's commitment problem, not by the AI's trustworthiness.
Schelling also supplies the corpus's own limiting condition (L582): deterrence has a documented advantage against state-like actors with survival stakes — a lab, a syndicate, a state — and a much weaker case against an actor indifferent to its own survival, such as a superintelligence that has copied itself. The corpus states this limit plainly, which is to its credit.
⚠ not retrieved — cited from memory and flagged. The transfer from interstate deterrence to machine deterrence is the discussion's inference, not Schelling's claim, and the vault records it as such.