Two Hidden Premises of Safety Institutions: Passivity and a Shared Substrate

Counts differ because each database indexes a different set of publications. We treat OpenAlex as the canonical count; Google Scholar is not shown (no API, and crawling it violates its ToS).

The institutions by which humans manage high-risk technologies and high-risk conflicts were formed through long practice. Engineering and medicine have lifecycle management, digital twins and traceable records; international conflict has mutual cost, third-party mediation and moral restraint. This paper argues that each of these two families rests on a premise that is usually left unstated. The first is passivity: the false-alarm tolerance and the response time in the engineering safety criterion can be chosen independently by the observer only because the observed object does not adjust itself to the process of observation. The second is a shared substrate: mutual cost requires the two sides to share a material substrate, third-party mediation requires both sides and the mediator to share an interest substrate, and moral restraint requires the two sides to share a biological substrate. We argue that AGI satisfies neither premise. On the side of passivity, we reinterpret the three core tools of engineering safety as three serial safeguards on a single chain whose purpose is to recognise a problem before hitting the wall, explain why they are dismantled at once, and why this is not a matter of insufficient refinement. The deceptive alignment behaviour already observed in the laboratory removes not the accuracy of detection but the ownership of two parameters of the engineering criterion. On the side of a shared substrate, we show the common source of three conflict buffers, and on that basis reduce the "mediation vacuum" to a structural gap that has always existed but has never been exploited: humanity as a whole has no peer. The two arguments are mutually independent; either one suffices to deprive a whole family of existing institutions of its purchase. We close with the few things both premises point to that can still be done — verifiable measurement and disclosure of capability, and the separation of capability from permission — and state the limits of each.