Framing safety as accidents, not superintelligence — polices a different boundary than → Drawing the boundary: what accident risk is not

Both exclusions surface in the paper's Section 8 survey of related efforts, but they exclude different kinds of things. The Futurist Community entry, source of long-term-ai-risk-framing, distances accidents from a rival framing of the same underlying concern: MIRI is credited with having studied several of this paper's own concepts, "wireheading, environmental embedding, counterfactual reasoning," "albeit at a more philosophical level," before the paper positions its empirical, near-term focus as complementary rather than opposed (concrete-problems, §"Futurist Community:", p. 20). The Related Problems in Safety entry, source of adjacent-research-areas' six fields, excludes not a rival framing but rival subject matter: privacy, fairness, security, abuse, transparency, and policy name different harms entirely, discrimination, adversarial attack, data exposure, that a system could commit while causing no accident at all. The paper's self-scoping thus runs along two independent axes: how to talk about accident risk, and what counts as accident risk in the first place.