Conjecture

There is a third place the correction loop can close, and it relocates the constituency rather than removing it: mutual exposure among the producers themselves, where each actor bears the cost of every other actor's failure. The industry becomes the harmed party. This needs no protesting victim and no imposed liability, and it explains the one case usually offered as culture rather than incentive. Aviation shares safety findings between competitors because a crash at one carrier damages public trust in flying, so best practices are "not corporate secrets, an edge over rival airlines" but "enthusiastically implemented by competitors in the common interests of collective industry trust and safety". The structural condition is that failure is attributed to the category rather than to the firm. NARROWED on review: this is weaker than the Shiwa dike case it was first assembled from. At Shiwa a farmer is exposed to one identified other farmer's infraction, immediately; airlines share a reputational pool that is diffuse, delayed and nobody's in particular, which is the channel that elsewhere in this neighborhood decays into ceremony. So the route does not escape the standing requirement. The test for AI therefore needs a clock, not just a direction: whether an incident at one lab imposes cost on its rivals ENOUGH, AND SOON ENOUGH, to be worth acting on before the incident, rather than merely transferring advantage to them. Current evidence is mixed and leans against. OpenAI's own incident lead framed the gap at industry level, "we are not there as an industry", which is the aviation register; but an outside researcher concluded "there has to be government oversight, because it won't happen otherwise", which is the opposite diagnosis, and Zvi records "a rather severe missing mood" in the response itself.

Institutional design